DNA replication begins at specific genomic locations known as replication origins. From each origin, molecular machines assemble and copy the surrounding chromosome in both directions. Bacteria often use a single origin, but eukaryotic chromosomes are much larger and typically require many starting points. The researchers focused on Saccharomyces cerevisiae, or budding yeast, whose replication origins are unusually well characterized. In this organism, origins occur within autonomous replication sequences, or ARS regions, generally about 100 to 200 base pairs long. These regions contain several functional elements, including the A element and B1, which together help form the principal binding site for the origin recognition complex, or ORC. The B2 element may provide an additional ORC-binding site or a platform for components of the replicative helicase.

A central sequence signal in yeast origins is the 11-base-pair ARS consensus sequence, or ACS. It is commonly represented as WTTTAYRTTTW, where W means either adenine or thymine and Y means cytosine or thymine. An extended version spans 17 base pairs and is written WWW-WTTTAYRTTTW-GTT. Yet the ACS alone cannot explain which origins actually function. The yeast genome contains roughly 12,000 matches to ACS-like motifs, but only about 500 are functional under normal conditions. Origin activity is therefore shaped by additional factors, including chromatin accessibility, transcription that can interfere with initiation, and secondary DNA features. This mismatch between abundant sequence motifs and comparatively rare active origins makes replication-origin prediction a demanding test for machine learning.

Earlier computational approaches often depended on labor-intensive feature engineering. Researchers had to convert sequences into numerical descriptions of nucleotide composition, DNA shape, physical properties, or sequence order before training support-vector machines, random forests, or other classifiers. Deep-learning methods reduced some of that manual work but frequently required models to be trained from scratch, and their learned features could be difficult to connect to recognizable biological mechanisms. The new study instead used pretrained transformer models. Transformers process sequences through self-attention, a mechanism that allows each input segment to weigh information from other segments while constructing a contextual representation. In principle, this enables a model to detect both short motifs and relationships between separated regions without being told in advance which biological features to search for.

DNABERT and DNABERT-2 share the basic BERT architecture, but they read DNA differently. DNABERT was pretrained on the human genome and converts sequences into overlapping four-base units, or 4-mers. Because adjacent tokens overlap, a DNA sequence is represented through a dense series of partially shared local windows. DNABERT-2 was pretrained on genomes from multiple species, including yeast, and uses byte-pair encoding, or BPE. BPE builds a vocabulary by repeatedly merging frequently occurring sequence segments, producing tokens of variable length. The newer model has about 117 million parameters and a vocabulary of 4,096 tokens, compared with approximately 86 million parameters for DNABERT. The difference in size comes mainly from the larger embedding matrix needed for BPE, not from deeper or wider transformer layers.

To train and test the models, the team assembled balanced datasets using 325 experimentally confirmed yeast origins from the curated OriDB resource. Each origin sequence was placed into a standardized 500-base-pair window. Shorter origins were extended with genuine neighboring genomic DNA, with the extra sequence distributed randomly between the two sides. This prevented the model from learning a trivial rule based on where the origin appeared within the window. The investigators created two contrasting negative datasets. In the Random-Neg set, non-origin sequences were sampled from genomic regions that did not overlap known origins. In the more difficult ACS-Neg set, negative sequences were selected from approximately 12,000 ACS matches that do not function as origins. Positive and negative sequences in the latter set therefore contained ACS-like signals of comparable strength, forcing the models to search for additional distinctions.

Both models were fine-tuned for binary classification: origin-containing sequence versus non-origin sequence. The DNA entered each model with special beginning and end tokens, while DNABERT-2 sequences were padded when its variable-length tokenization required equal input lengths. A classification layer attached to the representation of the initial classification token produced the probability that a sequence belonged to the origin class. The researchers used seven independent 70/10/20 training, validation, and test splits, selecting checkpoints based on validation accuracy and area under the receiver operating characteristic curve rather than simply taking the final training epoch. This design was intended to test robustness rather than maximize a single headline score. They also performed chromosome-based splitting to check that random partitioning had not artificially inflated performance.

DNABERT achieved an average test accuracy of 0.83 and an area under the curve of 0.90 on the easier Random-Neg dataset. DNABERT-2 reached 0.81 accuracy and 0.82 area under the curve. When ACS-rich non-origins were used as negatives, both models achieved approximately 0.72 accuracy, demonstrating that the task became substantially harder once the most obvious motif-based distinction was removed. The results indicate that the models were not merely detecting the presence of an ACS match. They retained some ability to distinguish functional origins from nonfunctional ACS sites, although the moderate accuracy also shows that sequence alone does not fully determine origin activity. Chromatin state, transcriptional context, and other cellular features remain outside the information available in the sequence windows.

The most revealing differences appeared when the researchers examined how the models arrived at their decisions. For DNABERT, they extracted attention scores associated with the classification token and projected those token-level values back onto individual nucleotides. Sharp attention peaks repeatedly fell within experimentally annotated origin regions rather than being distributed randomly across the 500-base-pair windows. The team collected short, 20-base-pair fragments around the strongest peaks and analyzed them with MEME, a probabilistic motif-discovery program that uses expectation-maximization to align variable motif instances. In test sequences from the Random-Neg dataset, the resulting motif was TTTTTWTTTATRTTT, with an E-value of 2.6 × 10−6, closely matching the known ACS pattern. A related training-set motif, TATATTTATRTWTWT, had an E-value of 2.3 × 10−32. In the ACS-Neg condition, where ACS motifs appeared in both classes, no comparably significant test-set motif emerged, consistent with the greater difficulty of the classification problem.

Attention maps from DNABERT-2 were less straightforward to interpret, so the researchers used perturbation experiments and Shapley additive explanations, or SHAP. Perturbation analysis asks what happens when a portion of the input is removed, altered, or rearranged. Eliminating ACS motifs from origin sequences changed DNABERT-2’s predictions, confirming that the motif contributed to classification. Randomly shuffling the model’s BPE tokens produced another important result: token identity appeared to matter strongly, while the precise order of some tokens mattered less than expected. SHAP analysis, repeated 30 times to reduce the randomness of its estimates, identified tokens that consistently supported origin or non-origin predictions. The team combined these findings into an AT-index incorporating overall adenine-thymine content, frequencies of AT-rich motifs, the longest uninterrupted AT run, and alternating AT runs. The index provided a compact way to quantify the AT-rich character of sequences associated with the model’s decisions, but the authors caution that SHAP values describe model behavior relative to its training background, not universal causal rules governing DNA replication.

The study also used shuffled sequences to test whether the models depended mainly on nucleotide composition or on the arrangement of bases. In one control, nucleotides within each positive sequence were randomly rearranged, preserving overall composition but destroying sequence order. In another, the sequence was divided into five-base-pair blocks and the blocks were shuffled, preserving local patterns while disrupting larger-scale organization. These experiments were designed to reveal whether the classifiers had learned broad compositional differences between origins and non-origins or more structured sequence information. The models’ near-random performance on shuffled data before task-specific fine-tuning showed that pretraining alone did not automatically confer the ability to distinguish intact from rearranged origin sequences. That capability emerged during adaptation to the replication-origin task.

Training from scratch produced weaker results than fine-tuning the pretrained systems. On the Random-Neg dataset, DNABERT trained from scratch averaged 0.71 accuracy, while DNABERT-2 averaged 0.60. The pretrained models used without fine-tuning performed at 0.37 and 0.55 accuracy, respectively, showing that general DNA representations alone were insufficient. Fine-tuning was therefore essential, but pretraining still supplied a useful starting point. The researchers emphasize that their goal was not to declare one architecture the universal winner. Instead, they wanted to determine whether genomic language models could reduce manual feature engineering while producing biologically interpretable signals. On that measure, the two systems behaved differently: overlapping k-mers gave DNABERT attention maps that more visibly reflected known motifs, whereas BPE tokenization in DNABERT-2 produced less transparent attention but supported alternative attribution analyses.

The findings matter because replication origins are not simply strings containing one magic sequence. In yeast, the ACS is an important anchor for ORC recognition, yet thousands of similar matches fail to initiate replication. A useful predictive model must therefore identify combinations of sequence properties while avoiding shortcuts created by biased negative examples. By deliberately challenging the models with ACS-containing non-origins, the researchers showed that transformer classifiers can recover signals beyond the canonical motif, though not perfectly. The work also illustrates a broader principle for biological AI: predictive accuracy is only part of the result. Tokenization, dataset construction, and explanation method can substantially influence what a model appears to have learned. As genomic language models are applied to human and other eukaryotic genomes—where replication origins are less sharply defined—the ability to distinguish meaningful biological signals from statistical artifacts will be crucial.

Dr. Law’s research centers on the complex chemical modifications known as epigenetic marks—molecular tags that dictate gene expression and maintain genome integrity without altering the underlying DNA sequence. These modifications play crucial roles in development, genome stability, and environmental response. By dissecting the biosynthetic and regulatory pathways that establish and remodel epigenetic landscapes, she illuminates fundamental principles that govern cellular function across life forms. Her model organism of choice, Arabidopsis thaliana, offers a robust platform for dissecting epigenetic phenomena, given that disruptions in such pathways in animals tend to be lethal.

One of Dr. Law’s landmark achievements involves revealing how transcription factors, alongside intrinsic DNA sequences, can orchestrate the establishment of new epigenetic patterns. This discovery advances the understanding of epigenetic regulation from passive modification to an active, targeted process, reshaping paradigms in genome biology. Additionally, her laboratory explores how chromatin architecture influences DNA repair dynamics, connecting epigenetic regulation to genome maintenance. This insight holds promising implications for bioengineering plants that can better endure environmental stressors, thereby supporting agricultural resilience and sustainability.

Dr. Law’s affiliation as a Rita Allen Foundation Scholar underscores her rising influence in the field. Her involvement with the Harnessing Plants Initiative at Salk further emphasizes her commitment to translating epigenetic insights into tangible agricultural innovations aimed at enhancing carbon sequestration and crop robustness under climate stress. This translational vision bridges molecular biology with global ecological and food security challenges, reflecting the broader impact of epigenetic research.

On the other side of the research spectrum, Dr. Talmo Pereira integrates computational neuroscience and artificial intelligence to decode the mechanics of biological movement. His work investigates how living organisms—from plants to humans—have evolved intricate locomotion strategies as survival mechanisms, and how these movements can serve as proxies to understand underlying neural processes. By constructing virtual simulations that mirror real-life behaviors, he probes questions of brain function and disease onset, particularly through the analysis of nuanced body language data.

Pereira’s past achievements include the development of SLEAP, an AI-driven, open-source software tool designed for markerless motion capture. This technology enables researchers worldwide to track and analyze movement patterns without traditional tagging methods, offering unprecedented accessibility and precision. Its adoption by tens of thousands of users globally highlights its utility across diverse species, from subcellular components to majestic whale sharks, underscoring the software’s wide-ranging applicability in biological research.

Looking ahead, Dr. Pereira’s ambitious projects aim to construct “embodied digital twins”—detailed virtual replicas of living animals. These models promise to advance our understanding of the neural codes that govern motor outputs, bridging behavior and brain activity with computational insight. Such integration of AI and neuroscience heralds new frontiers in personalized disease diagnostics and therapeutic developments based on movement biomarkers.

Gerald Joyce, MD, PhD, President of the Salk Institute, extols the creativity and boldness embodied by both Drs. Law and Pereira. Their research exemplifies Salk’s ethos of curiosity-driven science that not only addresses fundamental biological questions but also engenders innovations with broad societal implications. As these scientists push the boundaries of their fields, their discoveries promise to catalyze transformative advancements across genetics, computational biology, and beyond.

The significance of epigenetics in contemporary biology cannot be overstated. Modifications such as DNA methylation and histone modifications dictate cellular identity and adaptability, and dysregulation often precipitates pathological states. Dr. Law’s elucidation of how epigenetic patterns are dynamically established during development challenges previous assumptions of epigenetic marks as static and irreversible. Her findings enrich our comprehension of developmental plasticity and pave the way for engineering strategies that modulate genome function for crop improvement and stress tolerance.

Meanwhile, the fusion of computational tools with biological inquiry exemplified by Dr. Pereira’s research underscores the interdisciplinary shift transforming life sciences. The ability to capture, analyze, and simulate complex biological movements via AI platforms like SLEAP propels new understandings of behavior, neurobiology, and disease phenotyping. These methodologies could fundamentally alter how researchers approach diagnostics and therapy, creating personalized biomedical paradigms grounded in precise movement data analytics.

Together, the elevation of Dr. Law and the addition of Dr. Pereira to the Salk faculty represent a reinforcement of the Institute’s commitment to pioneering research at the nexus of molecular biology, genetics, computational science, and artificial intelligence. Their work exemplifies how foundational discoveries in plant epigenetics and computational neuroscience have the potential to transcend disciplinary boundaries and contribute solutions to pressing global challenges, including agricultural sustainability, environmental resilience, and human health.

Founded in 1960 by Jonas Salk, the Salk Institute remains a premier independent research institution that fosters collaborative and high-risk-high-reward scientific endeavors. Its mission continues to inspire discoveries that illuminate biological mysteries and translate into medical and technological breakthroughs. With faculty such as Drs. Law and Pereira, the Institute strides confidently toward the future, charting new paths in science that promise to benefit society profoundly.

As research advances rapidly in fields touching genetics, plant biology, computational neuroscience, and AI, the prominence of scientists who blend deep domain expertise with innovative technological approaches becomes increasingly vital. The careers and contributions of Julie Law and Talmo Pereira epitomize this fusion, symbolizing a new generation of researchers whose curiosity and rigor stimulate progress in understanding life’s complexity at multiple scales.

The ongoing exploration of epigenetic regulation and computational modeling of life’s movements at Salk is emblematic of the Institute’s role as an epicenter of discovery. These endeavors herald scientific transformations that will not only deepen our grasp of biological systems but also inspire novel interventions to address biodiversity, environmental stewardship, and health challenges worldwide.

Subject of Research: Epigenetic regulation in plants and computational neuroscience applied to biological movement analysis.

Article Title: Salk Institute Advances Epigenetics and Artificial Intelligence Research with New Faculty Appointments

News Publication Date: May 21, 2026

Web References: https://www.salk.edu/scientist/julie-law/; https://www.salk.edu/scientist/talmo-pereira/; http://www.salk.edu/

Image Credits: Salk Institute

Keywords: Epigenetics, DNA methylation, transcription factors, Arabidopsis thaliana, chromatin, DNA repair, computational neuroscience, artificial intelligence, motion capture, SLEAP, embodied digital twins, plant sciences, genome regulation.

The research indicates that traditional methods of studying drug responses often fall short due to their inability to comprehensively account for the multifaceted nature of cellular systems. The authors of the study highlight the limitations of previous models, which often either oversimplify biological processes or lack the computational power necessary to accurately predict outcomes. This new dual-branch transformer model addresses these shortcomings by integrating biological principles with advanced machine learning techniques, enabling a more nuanced analysis of cellular responses to drug treatments.

At the heart of this innovation is the dual-branch structure of the transformer model, which marries graph-based learning with sequence-based approaches. The researchers harness the potential of graph neural networks to capture the complex networks of interactions between proteins, genes, and other cellular components during drug exposure. This is complemented by transformer architectures that specialize in natural language processing, facilitating the modeling of temporal dynamics and the sequential nature of biological events. The synergy of these two methodologies presents a sophisticated tool for predicting how cells will react to various therapeutic agents.

The implications of this research extend far beyond mere academic interest. As the pharmaceutical industry increasingly relies on complex data for drug discovery and development, the ability to accurately model cellular responses could significantly streamline the process. By better understanding how different drugs elicit specific cellular reactions, researchers can optimize drug formulations and tailor therapies to individual patient profiles. This move towards personalization in medicine could not only enhance efficacy but also minimize adverse effects, ultimately improving patient outcomes.

In addition to its immediate applicability in drug development, the dual-branch transformer model holds potential for broader applications in pharmacovigilance and drug repurposing. The ability to swiftly and accurately assess how drugs impact cellular systems could aid regulatory bodies in evaluating the safety profiles of existing medications. Furthermore, the knowledge gained from cellular perturbation modeling may uncover new uses for well-established drugs, providing an expeditious path to novel treatments without the need for full re-approval processes.

The study employed extensive datasets derived from various cellular perturbation experiments to train the dual-branch model. This comprehensive approach not only reinforced the robustness of the findings but also elucidated patterns that might be overlooked by conventional analysis methods. The researchers meticulously analyzed the input features that define the model, ensuring that both biological relevance and computational efficacy are maintained. This dual emphasis on science and technology exemplifies the future of biomedical research, where interdisciplinary collaboration drives innovation.

As the dual-branch transformer model gains traction, further research will be needed to validate its predictive capabilities across diverse biological contexts. For instance, ongoing studies are expected to incorporate a wider array of cell types, drug classes, and perturbation experiments. These efforts will culminate in a more holistic understanding of cellular dynamics, and the hope is that this advancing knowledge will prompt systematic changes in how drugs are tested and approved.

In a climate where collaborative efforts lead to exponential gains in knowledge, the partnership between data scientists, biologists, and pharmacologists will be essential to fully leverage the potential of the dual-branch transformer. This holistic approach could serve as a blueprint for future innovations in the biosciences, drawing upon the strengths of multiple disciplines to create more comprehensive and accurate predictive models.

The authors emphasize that by continuously refining the model’s parameters with new data, the adaptability of the dual-branch transformer will be instrumental in accommodating the ever-evolving landscape of drug discovery. This dynamic nature allows the model not only to improve its performance over time but also to remain relevant as new therapeutic modalities emerge.

Moreover, the researchers anticipate that the model will soon be integrated into virtual environments that simulate drug interactions in silico. Such platforms could dramatically reduce the time and cost associated with preclinical studies, paving the way for more rapid advancements in translational medicine. As a result, researchers may be able to bring promising therapies from the laboratory bench to the bedside with unprecedented efficiency.

Finally, the study represents a significant stepping stone in the ongoing dialogue about the role of artificial intelligence in the life sciences. The dual-branch transformer model is not just a technological feat; it embodies a philosophical shift towards embracing computational tools as vital partners in biological research. As we continue to explore the complexities of life at the cellular level, it is crucial that we remain open to innovative methodologies that enhance our understanding and shape the future of healthcare.

In summation, the research led by Guo, Zhang, and Hu provides an exciting glimpse into the future of drug interaction modeling. The dual-branch transformer model stands to transform our approach to cellular perturbation responses, reinforcing the potential for enhanced therapeutic interventions in the realm of precision medicine. As this transformative technology finds its footing within the scientific community, it promises to enhance not only our understanding of pharmacodynamics but also the very nature of drug discovery itself.

Subject of Research: Drug-induced cellular perturbation responses using a dual-branch transformer model

Article Title: Modelling drug-induced cellular perturbation responses with a biologically informed dual-branch transformer.

Article References:
Guo, Y., Zhang, H., Hu, H. et al. Modelling drug-induced cellular perturbation responses with a biologically informed dual-branch transformer.
Nat Mach Intell (2026). https://doi.org/10.1038/s42256-025-01165-w

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-025-01165-w

Keywords: Drug discovery, cellular perturbation, dual-branch transformer, machine learning, precision medicine.

Cell–cell interactions are crucial for the development and proper functioning of tissues and organs. At the molecular level, these interactions are mediated through signaling pathways involving ligand–receptor pairs, which are influenced by the spatial arrangement of cells. Traditional methods of studying these interactions have often been hampered by limited abilities to measure ligand–receptor pairs, resulting in a fragmented understanding of their roles in various biological contexts. The emergence of spatial transcriptomics has introduced a transformative dimension, allowing researchers to visualize cellular interactions with great detail.

Despite the promising capabilities of spatial transcriptomics, significant challenges remain. Current analysis approaches often struggle with insufficient spatial encoding, limiting the ability to accurately interpret the multifaceted nature of cell interactions. This gap in understanding is addressed by the self-supervised learning model known as GITIII, which conceptualizes cells not merely as isolated entities but as integral parts of a communicative network. Through innovative techniques, GITIII captures the contextual relationships that shape cellular behavior and gene expression.

At the heart of GITIII’s functionality is the conceptualization of cells as “words,” with their surrounding cellular milieu representing the “context” that influences their state. By examining the relationships between a cell’s state and the characteristics of its neighborhood, GITIII effectively infers CCI patterns and elucidates how signaling from sender cells impacts the genetic programming of receiver cells. This nuanced understanding is particularly important for dissecting the complexities of ecosystems such as the brain and tumor microenvironments, where localized cellular interactions dictate larger biological outcomes.

To demonstrate its efficacy, GITIII was applied to a diverse array of four spatial transcriptomics datasets encompassing multiple species, organs, and technological platforms. The results were striking; GITIII not only successfully identified CCI patterns but also provided statistically meaningful interpretations of these interactions. This capability is particularly relevant in understanding the cellular dynamics within the brain, where various cell types work in concert to facilitate cognitive functions, as well as in tumor microenvironments, where interactions can influence cancer progression and treatment responses.

The interpretability of GITIII is a key factor that distinguishes it from other models. Often, advanced AI techniques can operate like ‘black boxes,’ yielding results that lack transparency. However, GITIII’s architecture is designed to be interpretable, allowing researchers to glean insights into the mechanisms driving cellular interactions. This interpretability is vital for translating findings into clinical applications, where understanding the ‘how’ and ‘why’ behind CCI can lead to novel therapeutic strategies.

Furthermore, GITIII enables visualization of spatial CCI patterns, which offers an intuitive perspective of how cells communicate with one another in their natural habitat. This feature is particularly important for researchers striving to map the intricacies of tissues and understand how dysregulation of these interactions might contribute to disease states. By providing a clearer picture of spatially-driven cellular communication, GITIII serves as a powerful tool for both basic and translational research.

Additionally, GITIII’s capabilities extend to CCI-informed cell clustering, an analytical process that aids in categorizing cells based on their interaction profiles. This sophisticated clustering allows for a more refined understanding of cellular heterogeneity within tissues. In contexts such as cancer, where the tumor microenvironment is known to significantly influence treatment efficacy, understanding these clusters can provide insights into why certain therapies fail and guide the development of more effective strategies.

As GITIII continues to be tested across various datasets and biological systems, the implications of its findings are poised to impact a multitude of fields, from developmental biology to oncology. By facilitating a deeper understanding of how cells communicate, GITIII not only contributes to our basic scientific knowledge but also holds promise for addressing some of the most pressing challenges in medicine today.

In conclusion, the advent of GITIII represents a significant leap forward in the field of cell biology, particularly in how we approach the study of cell–cell interactions. Its innovative use of self-supervised learning, combined with a robust interpretative framework, provides researchers with new tools to decipher the complex web of communication that governs cellular behavior. With further validation and expansion, GITIII is positioned to become an essential asset in the ongoing quest to understand the intricacies of life at the cellular level.

This revolutionary approach illustrates the potential of integrating modern computational methods with biological research, paving the way for future advancements in understanding the fundamental processes that underpin living organisms. As the boundaries of cell biology are continuously pushed, tools like GITIII will play a critical role in shaping the future of biomedical research and improving patient outcomes.

Subject of Research: Cell–Cell Interactions and Spatial Transcriptomics

Article Title: Inferring spatial single-cell-level interactions through interpreting cell state and niche correlations learned by self-supervised graph transformer

Article References:
Xiao, X., Zhang, L., Zhao, H. et al. Inferring spatial single-cell-level interactions through interpreting cell state and niche correlations learned by self-supervised graph transformer.
Nat Mach Intell (2025). https://doi.org/10.1038/s42256-025-01161-0

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-025-01161-0

Keywords: Cell–Cell Interactions, Spatial Transcriptomics, Graph Transformer, Self-Supervised Learning, Cancer Microenvironments, Gene Expression, Interpretable AI.

The formal relationship underscores a shared commitment to fostering rapid, open dissemination of pioneering research in the fields of bioinformatics and computational biology. Through this union, MCBIOS members gain an invaluable platform to publish their work, notably research contributions presented at their annual conference. This ensures that innovative findings, particularly in artificial intelligence, machine learning, and big data analytics applied to biological systems, reach the global scientific audience with immediacy and through a rigorous peer-reviewed process.

JMIR Bioinformatics and Biotechnology, already recognized for rigorous editorial standards and a comprehensive open access model, now becomes the official journal of MCBIOS. This designation not only enhances the journal’s prestige but also guarantees a steady influx of high-impact research reflecting the latest technological and scientific advances in computational biology. The partnership facilitates a unified voice for MCBIOS researchers, providing an authoritative venue to broadcast their discoveries and methodologies.

One of the standout features of this partnership is the establishment of a dedicated annual conference theme issue, curated in collaboration with MCBIOS. This special issue will compile select research highlights from MCBIOS’s yearly scientific meetings, ensuring these works achieve wider indexing and recognition. Such systematic coverage fosters a scholarly repository that documents evolving trends and breakthroughs in computational biology and its interdisciplinary interfaces, strengthening the foundation for future collaborations.

Beyond publication avenues, MCBIOS members benefit directly from discounted Article Processing Fees (APFs) for accepted manuscripts in JMIR Bioinformatics and Biotechnology. This financial incentive lowers barriers for researchers to disseminate their work openly, encouraging a higher volume of submissions and promoting inclusivity among a diverse pool of scientists. The open access nature assures global dissemination, with no subscription barriers impeding the flow of scientific knowledge.

Complementing publication support, JMIR Publications commits to delivering virtual training programs aimed at enhancing members’ scholarly publishing proficiency. These educational initiatives address intricacies of manuscript preparation, open science principles, peer review navigation, and ethical considerations in academic publishing. By empowering MCBIOS members with these skills, the partnership actively cultivates a community of researchers adept at maximizing the impact and reach of their scientific contributions.

Dr. Aik Choon Tan, President of MCBIOS, emphasizes the strategic importance of this collaboration, noting that years of diligent efforts to secure a reliable, quality-focused publishing home have culminated in this milestone. He highlights how the partnership ensures that novel insights in computational biology, data-driven biological modeling, and biotechnology gain the rapid and open visibility they merit. JMIR’s commitment to open science aligns seamlessly with MCBIOS’s mission to elevate the mid-southern academic ecosystem.

JMIR Publications Vice President of Communications and Partnerships, Dennis O’Brien, remarks that the partnership represents a shared vision to revolutionize scholarly communication. By fostering a premier open access venue with dedicated features like the thematic conference issue and member discounts, JMIR actively contributes to the acceleration and amplification of bioinformatics research. This collaboration exemplifies a forward-looking model where technology-enabled publishing maximizes scientific impact and accessibility.

MCBIOS will exclusively endorse JMIR Bioinformatics and Biotechnology as the official publishing platform for its annual conference proceedings and other scholarly events. This endorsement encourages members to prioritize submitting their most impactful research to the journal, cementing the partnership’s role as a central hub for cutting-edge bioinformatics discourse. The resulting synergy is expected to increase citation traction, cross-disciplinary integration, and collaborative innovation.

The MidSouth Computational Biology and Bioinformatics Society has long prioritized advancing computational and integrative biology through education, community building, and scientific dissemination. This partnership with JMIR Publications strengthens these core pillars by furnishing members with enhanced publication opportunities, professional development, and international visibility. It positions the regional society as a critical node in the global bioinformatics network.

JMIR Publications stands at the forefront of open access digital health and biomedical publishing, driving innovation beyond traditional academic dissemination. Their portfolio, which includes highly indexed and prestigious journals, offers researchers a comprehensive suite of tools and resources designed to augment publication quality, outreach, and career growth. This strategic alliance with MCBIOS reinforces JMIR’s role as an enabler of scientific progress through accessible and impactful publishing.

In aligning editorial priorities, peer review excellence, open science principles, and community engagement, JMIR Bioinformatics and Biotechnology and MCBIOS set a new benchmark for professional society partnerships in the digital era. This ongoing collaboration is expected to catalyze novel discoveries in computational biology, foster robust scientific communication, and inspire further innovation at the intersection of biology, technology, and data science.

Subject of Research: Computational Biology, Bioinformatics, Artificial Intelligence in Biology, Big Data Analytics in Life Sciences

Article Title: JMIR Bioinformatics and Biotechnology Officially Designated as the Journal of The MidSouth Computational Biology and Bioinformatics Society

News Publication Date: November 11, 2025

Web References:
– JMIR Publications: https://jmirpublications.com/
– MidSouth Computational Biology and Bioinformatics Society: https://mcbios.com/
– JMIR Bioinformatics and Biotechnology Journal: https://bioinform.jmir.org/

Image Credits: JMIR Publications

Keywords: Biotechnology, Bioengineering, Biomedical Engineering, Medical Technology, Technology, Science Communication, Science Journalism, Academic Publishing, Academic Journals, Peer Review, Publishing Industry, Medical Journals, Scientific Journals

Catalysts are central to the chemical transformations that drive both biological processes and industrial manufacturing. Among catalysts, enzymes stand out due to their remarkable selectivity and efficiency, often outperforming synthetic alternatives under mild conditions. Yet, their inherent limitations—narrow operational environments and restricted substrate scope—present significant challenges. Natural enzymes are typically optimized for specific reactions within the confines of living systems, restricting their direct applicability in diverse synthetic contexts. Overcoming these barriers requires a paradigm shift toward designing enzymes that not only match but exceed natural capabilities in terms of stability, versatility, and reaction scope.

The research team tackled this challenge by employing a bottom-up strategy centered on de novo protein design, which constructs proteins purely from amino acid sequences without relying on existing natural templates. This approach leverages the modularity of amino acids to create minimalist yet highly functional protein frameworks, exemplified by simple helical bundle proteins. Such small, robust scaffolds offer advantages in thermal and solvent stability, tolerating conditions that would denature conventional enzymes. Moreover, these frameworks are amenable to incorporating unnatural cofactors and metal centers, broadening the catalytic repertoire beyond nature’s constraints.

To translate these design principles into functional catalysts, the collaborators applied cutting-edge artificial intelligence methods to predict amino acid sequences that would fold into proteins with the desired three-dimensional structures and reactive sites. This sequence optimization was coupled with in-house algorithms and crystallographic insights to iteratively refine the enzyme architecture. A pivotal moment in the process arose during X-ray crystallography analysis, revealing a disorganized loop region where a structured helix was intended. This structural imperfection underscored the complexity of enzyme design, indicating that AI predictions alone could not capture all subtle features critical for catalytic performance.

Addressing this, the team introduced a loop searching algorithm alongside expert chemical intuition to redesign and stabilize this region. The subsequent round of engineering drastically improved enzyme activity and stereoselectivity, with several variants demonstrating exceptionally high efficiency in catalyzing carbon-carbon and carbon-silicon bond formations. These reactions are of particular synthetic importance because natural enzymes that facilitate such transformations are scarce or inefficient. The success of these redesigned enzymes thus opens doors to new synthetic routes that are challenging or inaccessible through traditional bio- or chemo-catalysis.

This research embodies a fusion of computational innovation, structural biology, and synthetic chemistry, emphasizing that while AI accelerates design, human insight remains essential. The iterative cycle of prediction, validation, and refinement underscores a nuanced understanding of protein folding landscapes and active site dynamics. Such mastery enables the crafting of protein catalysts tailored for challenging transformations with precise control over stereochemical outcomes, an aspect crucial for the synthesis of complex molecules with pharmaceutical relevance.

A further breakthrough in this study is the ability to tune enzyme function by selecting cofactors that are rare or absent in nature. This flexibility allows chemists to exploit a palette of reactive centers to drive unique catalytic cycles, broadening the physicochemical parameters under which enzymes can operate. Notably, the proteins designed here maintain their catalytic activity in water—the greenest solvent available—aligning enzyme engineering efforts with sustainability goals and green chemistry principles.

Looking ahead, ongoing efforts by the Yang lab in close collaboration with the DeGrado and Liu labs focus on achieving simpler and smaller enzymes that rival or surpass complex natural enzymes in activity. Another ambitious goal is to design enzymes that catalyze reactions through mechanisms previously unknown in biological systems. If successful, this would profoundly expand the toolbox of chemical transformations accessible via biocatalysis and reshape industrial processes that currently rely heavily on environmentally intensive synthetic methods.

The implications of this work are far-reaching. Bespoke enzymes crafted for specific reactions could revolutionize drug discovery, enabling previously intractable synthetic routes to active pharmaceutical ingredients with fewer steps, higher selectivity, and less waste. In materials science, such catalysts could facilitate the assembly of novel polymers and advanced materials under mild conditions, reducing the carbon footprint of manufacturing. Moreover, by decoupling enzyme design from natural constraints, chemists gain access to a virtually limitless space of protein-based catalysts adapted to diverse applications.

This study reflects a significant milestone in enzyme engineering, demonstrating how interdisciplinary collaboration accelerates innovation at the intersection of biology, chemistry, and computational science. Its success also highlights that the journey to fully artificial enzymes demands not only sophisticated algorithms but also deep chemical understanding and precise experimental validation. The synergistic combination of these elements sets a new standard for rational enzyme design.

The team, including Kaipeng Hou, Wei Huang, Miao Qui, Thomas H. Tugwell, Turki Alturaifi, Yuda Chen, Xingjie Zhang, Lei Lu, and Samuel I. Mann, illustrates a new era where human-guided AI design catalyzes breakthroughs that are both scientifically profound and practically transformative. As this field progresses, it promises to make enzyme design an accessible and routine tool, democratizing the ability to tailor powerful catalysts for the sustainable technologies of tomorrow.

Subject of Research: De novo enzyme design and protein engineering for synthetic catalysis

Article Title: (Not specified in the original content)

News Publication Date: (Not specified in the original content)

Web References: https://www.science.org/doi/10.1126/science.adt7268

References: (Detailed references not provided in the original content)

Image Credits: (Not specified in the original content)

Keywords

Applied sciences and engineering, Enzyme design

The significance of the MetaVision3D lies in its ability to highlight the full spectrum of molecules that are integral to energy production within brain cells. This novel perspective allows for a deeper exploration into the biochemical landscape of the brain, potentially illuminating the metabolic pathways that may be altered in various disease states. The implications of such research are profound, offering new avenues for targeted therapeutic interventions aimed at metabolic dysregulation—a feature prominently associated with Alzheimer’s disease and other cognitive disorders.

Funded by the National Institutes of Health, the development of MetaVision3D represents a remarkable leap in the application of technology and artificial intelligence in neurobiological research. The framework enables researchers to create detailed, interactive atlases of both healthy and diseased brain states, enabling them to visualize, analyze, and ultimately comprehend how cellular metabolism interacts with brain function. The project is particularly timely, given the increasing urgency to understand the molecular underpinnings that contribute to complex diseases affecting millions globally.

At the heart of the project is Dr. Ramon Sun, a leading figure in the fields of spatial biomolecule research and neuroscience. Under his direction, the research team employed UF’s HiPerGator supercomputer to produce a remarkably detailed brain atlas. This endeavor was not merely an exercise in high-tech imaging; it was a meticulous process of layering—scanning 79 brain sections in minuscule increments to compile a comprehensive representation of the brain’s metabolome, the aggregate of molecules that fuel neural function. By employing advanced imaging techniques, the team was able to capture sensitive details of molecular architecture that previously eluded researchers using traditional two-dimensional imaging methods.

The reconstruction of this 3D metabolomic map involved utilizing sophisticated artificial intelligence tools to align and integrate the vast array of images collected throughout the scanning process. According to Dr. Xin Ma, a pivotal member of the research team and a doctoral student, this method allowed researchers to approximate the spatial organization and distribution of thousands of metabolites within the brain, achieving remarkable accuracy levels ranging from 95 to 99%. This exceptional precision is critical for developing reliable models that can elucidate the metabolic disruptions linked to neurodegenerative conditions.

The interactive nature of the MetaVision3D tool empowers users to engage with brain structures in ways previously thought unattainable. By offering the ability to zoom in on specific brain regions, researchers can visually dissect the intricate cellular processes playing out in real-time. This dynamic approach heralds a new era for scientists investigating the multifaceted relations between metabolism, cognition, and disease—a field that has greatly benefitted from advancements in biochemistry and artificial intelligence.

One of the unique features of the framework is its capacity to correlate anatomical structures with metabolic pathways. By mapping the metabolic landscape of the brain in both normal and disease states, the researchers hope to uncover the nuanced changes that occur during the progression of neurodegenerative diseases. For instance, understanding how specific molecules influence cognitive processes such as memory and learning may shed light on targets for therapeutic intervention. With traditional treatment methods often impacting both healthy and diseased tissue alike, the precision of this mapping tool could prove transformative in devising strategies that selectively target affected areas.

The potential of this technology extends beyond basic research, as it opens new possibilities for translational science. By integrating MetaVision3D with existing MRI imaging and genetic testing, researchers could pioneer new treatment paradigms that focus on localized interventions, thereby reducing unintended side effects. Dr. Sara Burke, another key investigator in the study, noted that such innovative approaches may well redefine the landscape of clinical neuroscience, shifting the paradigm towards more personalized and effective treatment approaches.

In closing, the arrival of MetaVision3D signals a pivotal shift in the methodological landscape of neuroscience. By combining high-resolution 3D mapping with AI-driven analysis, researchers now have access to a tool that may uncover critical insights into the biochemical foundations of brain health and disease. As work continues on this promising frontier, the scientific community eagerly anticipates the implications of these findings in shaping future therapeutic strategies for Alzheimer’s and other debilitating neurodegenerative conditions.

Furthermore, this pioneering research not only signifies an important step forward in our understanding of brain metabolism but also highlights the vital role that interdisciplinary collaboration plays in advancing scientific knowledge. With expertise from diverse fields coming together—from artificial intelligence to neuroscience—the potential to unlock the mysteries of the brain has never been greater. As the world grapples with rising rates of cognitive decline, innovations such as MetaVision3D serve as a beacon of hope in the search for efficacious treatments that could one day mitigate the impact of these devastating diseases on individuals and their families.

As we stand on the cusp of a new era in neurobiology, the excitement surrounding the MetaVision3D project is palpable. Researchers are optimistic that this advanced mapping tool will pave the way towards significant breakthroughs in understanding the interplay between metabolism and cognition, unlocking new methods to not only treat but potentially prevent neurodegenerative diseases before they establish a foothold. The journey of discovery continues, and with it, the promise of a future where brain health is better understood, and the devastating effects of cognitive decline are significantly reduced.

Subject of Research: Animals
Article Title: AI-driven framework to map the brain metabolome in three dimensions
News Publication Date: 18-Mar-2025
Web References: MetaVision3D Server
References: Nature Metabolism Paper
Image Credits: University of Florida

Keywords

Molecular mapping, Gene targeting, Molecular targets, Artificial Intelligence, Genetic mapping, Magnetic resonance imaging, Brain structure

The research identifies four primary categories of foundation models: language, vision, graph, and multimodal FMs. Each category offers unique strengths and capabilities, enhancing the approaches employed in genomics, transcriptomics, proteomics, drug discovery, and single-cell analysis. By systematically categorizing these models, the research provides a roadmap for scientists to select the most appropriate FM based on the specific needs of their bioinformatics applications. As new possibilities unfold, the intersection of AI and molecular biology emerges as a vibrant field ripe for exploration.

At the core of this study lies an emphasis on the versatility and adaptability of foundation models. Prof. Wang and his research team highlighted the potential of these models to be trained using both supervised and unsupervised learning techniques, making them ideal for addressing a spectrum of biological challenges. The research team underscored the necessity of integrating advanced AI technologies into the molecular biology workflow to create a robust framework for future innovations in the field. This integration not only improves predictive capabilities but also allows for a richer analysis of biological phenomena.

Central to this discourse is the understanding of biological databases, training strategies, and hyperparameter configurations, which are crucial components in the deployment of foundation models within bioinformatics. The team meticulously discussed how these attributes could be optimized to enhance performance and accuracy in various tasks. By providing insights into training techniques and configurations, the team lays the groundwork for future research, allowing others to build upon this knowledge and explore new avenues for discovery.

One of the standout contributions of this research is its focus on the evolutionary process of bioinformatics feature mapping. Through a comprehensive understanding of model advancements, the team elucidated how the improved models have mitigated the limitations faced by their predecessors. This evolutionary perspective emphasizes the ongoing nature of research in bioinformatics and highlights the significance of iterative development in crafting effective AI solutions for complex biological inquiries.

The discourse surrounding the practical applications of FMs reached a notable crescendo with the discussion surrounding DeepMind’s efforts in protein structure reconstruction. Prof. Wang pointed to DeepMind’s development of artificial intelligence systems over the past five years, showcasing the promising outcomes that have emerged from these advancements. By tying real-world applications to the evolving field of bioinformatics, the research feeds into a growing narrative of success and innovation, demonstrating the tangible impacts of AI on scientific research.

As the researchers prepared to publish their findings, they captured the broader implications of foundation models in bioinformatics. Their insights into model pre-training frameworks, benchmarking methods, and interpretability promises to shape the future direction of research in the field. The emphasis on model hallucination evaluation reflects a commitment to transparency and reliability, allowing other researchers to engage with these models while understanding their limitations.

Through their work, the team stresses the significance of comprehensive analysis and understanding of the underlying mechanisms that govern biological systems. This approach is essential, as it leads to new insights and generates hypotheses that advance our understanding of complex biological interactions. Additionally, the investigation of how models can interpret and generate biological data serves as a critical component in the continuous effort to bridge gaps in scientific knowledge.

The emerging narrative surrounding foundation models paints a picture of immense potential and possibility, inviting researchers from various domains to converge in interdisciplinary collaborations. By exploring the innovative capabilities of AI and combining expertise from computer science and biology, the research opens doors to novel methodologies that can revolutionize bioinformatics practices and contribute to advances in medicine and health.

Undoubtedly, the burgeoning field of bioinformatics stands at a pivotal juncture, with artificial intelligence-driven solutions poised to redefine biological research paradigms. As the research community rallies around these transformative technologies, the dialogue around foundation models serves as both a catalyst for collaboration and a testament to human ingenuity in the face of scientific challenges.

The team’s final thoughts resonate with a sense of optimism for the future of bioinformatics, as they highlight the potential for foundation models to not only elevate existing practices but also to catalyze groundbreaking discoveries. By charting this course, they aim to inspire a new generation of scientists to leverage AI technology in their quest to understand the intricacies of life itself.

In conclusion, this pivotal study encapsulates the essence of interdisciplinary collaboration and the transformative power of foundation models in bioinformatics. By highlighting theoretical underpinnings and practical applications, the researchers not only contribute to the academic landscape but also underscore the inevitability of AI’s central role in the future of biological research.

Subject of Research: Foundation Models in Bioinformatics
Article Title: Leveraging AI: The Future of Foundation Models in Bioinformatics
News Publication Date: October 2023
Web References: http://dx.doi.org/10.1093/nsr/nwaf028
References: National Science Review
Image Credits: ©Science China Press

Keywords: bioinformatics, foundation models, artificial intelligence, genomics, high-throughput data, molecular biology, protein structure, DeepMind, AI models, interdisciplinary research, biomedical applications.

At the heart of this research is the cone snail toxin known as Conkunitzin-S1 (Cs1). This toxin, primarily impacting potassium channels in the cells of fish and insects, poses a unique challenge for scientists seeking to understand its precise mechanisms of action. While it is well-documented that Cs1 effectively blocks potassium channels, rendering them unable to facilitate essential cellular functions, the specific targets within fish had remained elusive until now. Understanding these interactions is pivotal, not only for insights into ecological dynamics but also for drug development applications.

The research team, led by Izhar Karbat and Eitan Reuveny, faced significant hurdles when attempting to pinpoint the targets of Cs1 using conventional tools three years ago. Despite their best efforts, they could not attain the clarity required to establish a comprehensive understanding of the toxin’s interactions. However, the advent of advanced AI technologies has revolutionized their approach, enabling them to explore molecular interactions with unprecedented precision.

Utilizing the AI program AlphaFold, the scientists first predicted the binding interactions between Cs1 and an array of fish potassium channels. This step was crucial, as it provided a foundational understanding of which channels might be impacted by the toxin. By leveraging AlphaFold’s capabilities, Reuveny and Karbat laid the groundwork for a deeper analysis of these molecular interactions, enabling them to hypothesize how Cs1 engages with specific proteins.

In addition to using AlphaFold, the researchers developed ET3, an innovative AI model designed to analyze the dynamics of water molecules surrounding potassium channels. This model focuses on the selectivity filter—the part of the channel responsible for regulating ionic flux—understanding that disruptions in this region can lead to channel inactivation. ET3, trained on a wide assortment of potassium channels, excels at identifying anomalies in water movement, thereby illuminating potential binding sites for Cs1.

Through this dual approach, the research team was able to sift through a vast landscape of potassium channels previously unexplored by conventional methods. Their findings revealed the specific fish potassium channels that Cs1 targets, shedding light on the intricate dynamics of the toxin’s interaction. The research illustrates that Cs1 functions akin to a lock that seizes control of these ion gates, preventing potassium from traversing the channel.

Furthermore, Karbat expressed excitement over the broader applications of this research extending beyond the immediate ecological implications. The pipeline established through their work opens new avenues for drug discovery, offering a way to accurately determine the targets of newly developed drugs based on their structural characteristics. Such precision is particularly vital as it helps mitigate unintended side effects, such as a drug meant for brain channels inadvertently affecting cardiac channels.

This research also highlights the importance of understanding off-target interactions, especially in therapeutic contexts. For instance, if a drug developed to stimulate a potassium channel in neuronal tissues also activates similar channels in cardiac tissues, the consequences could be severe. Thus, the ability to accurately identify and differentiate targets presents a crucial stride in ensuring drug safety.

Moreover, the implications of the findings extend into ecological studies. By employing the newfound understanding of molecular interactions, researchers can delve deeper into ecological systems and the roles played by various toxins within them. This could lead to insights about how these interactions affect populations, ecosystems, and ultimately, biodiversity and conservation efforts.

In conclusion, the Weizmann Institute team’s innovative blend of artificial intelligence and traditional methods has marked a significant milestone in the field of molecular biology. The research not only enriches our understanding of cone snail toxins but also showcases the potential for AI to transform drug development strategies and ecological research. As scientists continue to unravel the complexities of molecular interactions, this work stands as a testament to the power of interdisciplinary strategies in advancing scientific inquiry.

Subject of Research: Interactions of Cone Snail Toxin with Potassium Channels
Article Title: Weizmann Institute Scientists Unravel Potassium Channel Interactions of Cone Snail Toxin Using AI
News Publication Date: TBD
Web References: TBD
References: TBD
Image Credits: Courtesy of Eitan Reuveny and Izhar Karbat
Keywords: Biophysics, Toxins, Molecular Biology, Artificial Intelligence, Drug Development