The appeal of using light as a regulatory input is easy to understand. Unlike drugs or small molecules, light can be applied instantly, confined to a specific region, and tuned in intensity without chemically perturbing the cell. Optogenetic circuits built from photosensitive proteins have transformed neuroscience, allowing researchers to map and manipulate neural activity with millisecond precision. But most of these circuits operate at the level of transcription, controlling whether a gene is read into messenger RNA in the first place. Translation, the step in which ribosomes convert mRNA into protein, offers an even more direct and immediate handle on protein levels, and it is the step that cells themselves regulate most heavily during stress, differentiation, and disease.

In living cells, that translational control is largely exerted by RNA-binding proteins, or RBPs, which latch onto specific sequences in messenger RNAs and either block or boost their translation. During epithelial-mesenchymal transition, for example, the RNA-binding protein RBFOX2 is enabled while ESRP1 is disabled, collectively raising the output of mesenchymal-related mRNAs and suppressing epithelial-related ones. Mimicking this kind of opposing, coordinated regulation with a synthetic device has been the missing piece. The HKUST team, led by Yi Kuang, set out to build a platform in which a single light input would activate one RNA-binding protein while simultaneously preventing the formation of another, allowing two different target mRNAs to be regulated in opposite directions at once.

The core of the design is a clever inversion of a standard optogenetic trick. The researchers began with a split intein, a pair of protein fragments derived from the DnaE intein of the cyanobacterium Nostoc punctiforme that spontaneously stitch themselves together and splice out, fusing whatever proteins are attached to their ends. Normally, light-induced dimerization is used to bring protein fragments together. Here, the team did the opposite: they fused one half of the intein to the blue-light dimerizing protein pMag, and the other half to its partner nMag, but inserted a rigid synthetic alpha-helical spacer of about 39.7 angstroms between the intein fragment and pMag. In the dark, the intein halves find each other and splice normally. Under blue light, pMag and nMag dimerize and physically drag the intein halves apart, holding them beyond the reach of the spacer and blocking splicing altogether.

The team validated the concept in HEK293T cells using fluorescent reporters. When cells expressing the two fusion proteins were kept in the dark, the fluorescent signals swapped localization, confirming that the intein had spliced and conjugated the proteins as intended. Western blot analysis sharpened the picture: dark-incubated cells showed a dominant band at 18.3 kilodaltons, the expected spliced product, accounting for 67.6 percent of the signal, while light-exposed cells showed a dominant band at 36.4 kilodaltons, the unspliced precursor, at 71.4 percent. Removing the spacer abolished the light response entirely, with splicing proceeding under both conditions, proving that the rigid helix was the element converting photodimerization into inhibition. The chosen light intensity of roughly 3.75 milliwatts per square centimeter caused no measurable loss of cell viability.

Crucially, the Light-OFF system proved orthogonal to an existing Light-ON intein system based on the light-oxygen-voltage (LOV) sensing domain, which assembles a different engineered intein pair, NpuM, under the same blue light. When both systems were placed in the same cells, one driving reconstitution of split sfGFP under light and the other driving reconstitution of split mCherry in the dark, the two fluorescent outputs cleanly inverted between conditions. sfGFP signal under light reached levels about 4.5-fold higher than in the dark, while mCherry in the dark was roughly twice the light-condition level. The researchers also showed that the Light-OFF mCherry output could be cycled: switching light and dark every 24 hours over five days made the mCherry signal repeatedly appear and disappear without harming the cells.

With the light-sensing machinery in hand, the team turned to the harder problem of engineering the RNA-binding proteins themselves. They chose two widely used model RBPs, the MS2 bacteriophage coat protein (MCP) and the PP7 coat protein (PCP), which bind distinct RNA aptamers and are workhorses for tracking and manipulating RNA in living cells. When bound to aptamers placed in the 5-prime untranslated region of a reporter mRNA, these proteins suppress translation. The challenge was that split inteins prefer particular amino acids at the splicing junction, and the flexible random-coil regions of MCP and PCP did not naturally offer suitable split sites. The researchers inserted cysteine residues at engineered positions and used AlphaFold 3 to simulate whether each candidate split pair would reconstitute into a structure resembling the original protein, screening candidates by template modeling score and root-mean-square deviation before ever testing them in cells.

The simulations paid off. Split MCP with a cysteine inserted at the 38/39 position and split PCP split at 36/37 proved the most potent, each reconstituting into functional repressors that bound their cognate aptamers and suppressed reporter expression. The team then assembled the full bidirectional photoriboswitch: light induces formation of PCP, which suppresses an EGFP reporter bearing the PP7 aptamer, while darkness allows formation of MCP, which suppresses an iRFP reporter bearing the MS2 aptamer. In cells carrying the full device, the EGFP-to-iRFP ratio shifted by more than 4.3-fold between light and dark conditions, and the platform performed similarly in HeLa cells, with a 6.7-fold shift, demonstrating that the effect is not confined to a single cell line.

Perhaps most strikingly, the platform could be flipped from repression to activation. Many natural RNA-binding proteins enhance translation rather than block it, and the team mimicked this by fusing the VPg translation-promoter motif, derived from a viral protein genome-linked factor, onto the split RBP fragments, while fitting the reporter mRNAs with a translation-deficient cap. Now, light-induced formation of PCP-VPg localized the enhancer onto the EGFP switch and boosted its output, while darkness enabled MCP-VPg to upregulate the iRFP switch. The light-to-dark ratio shift in this configuration exceeded 81.7-fold, a dramatic demonstration that the same architectural principle can be adapted to opposing modes of translational control simply by swapping the functional domain attached to the reconstituted protein.

The authors are candid about the current limitations. Blue light, the input used throughout the study, penetrates tissue poorly and can cause DNA damage, so extending the platform to red or near-infrared sensing pairs such as BphP1/QPAS1 will be essential for any move beyond cultured cells. The system also produces irreversible RBP formation, since intein splicing cannot be undone, although natural protein degradation and dilution during cell growth limit the duration of the effect and allow repeated light-dark cycles of regulation. And the demonstration so far regulates two model reporter mRNAs; extending the approach to multiple physiologically relevant transcripts remains future work. Even so, the design principles on display, using reversible photodimerization to gate irreversible protein assembly and using computational structure prediction to engineer split points where none naturally exist, offer a genuinely expandable foundation. As RNA-based therapeutics and synthetic mRNA circuits mature, tools that can dial protein production up and down with nothing more than light are likely to find eager users, from basic researchers dissecting gene regulation to engineers building cell therapies that respond to optical commands.

Subject of Research: A bidirectional light-controlled riboswitch for regulating translation of synthetic mRNAs in mammalian cells

Article Title: Bidirectional photoriboswitch for translational regulation in mammalian cells

Article References: Hu, Y., Li, C. Y., Fu, L., Sun, Y., Zhang, M., Yau, T. M., Xiong, C., Shi, P., & Kuang, Y. (2026). Bidirectional photoriboswitch for translational regulation in mammalian cells. iScience, 29(10), Article 117671. https://doi.org/10.1016/j.isci.2026.117671

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117671

Keywords: synthetic biology, optogenetics, photoriboswitch, translation regulation, RNA-binding proteins, split intein, mRNA, blue light, AlphaFold 3, mammalian cells, gene expression control, HKUST

The core advantage AI brings to oncology is its ability to digest high-dimensional, multi-omic data at a scale no human team could match. By integrating genomics, proteomics, and clinical trial data, machine learning algorithms can detect subtle, non-linear relationships that traditional statistical tools simply miss. The review highlights how AI-driven platforms have compressed the journey from target identification to lead compound optimization, potentially shrinking a process that once took years into a matter of months. That acceleration does more than save money; it enables the rapid identification of patient subgroups most likely to respond to a given therapy, which is the practical heart of precision medicine. Techniques spanning machine learning, deep learning, natural language processing, and reinforcement learning now touch every stage of the oncology drug lifecycle, from discovery benches to post-marketing surveillance.

Among the technical standouts is DeepDTA, a model built on convolutional neural networks that predicts binding affinities between drug compounds and target proteins using nothing more than sequence information. Benchmarked on the widely used KIBA and Davis datasets, it has outperformed traditional molecular docking approaches on metrics like mean squared error and concordance index. Equally intriguing is the Cascade Deep Forest model, an ensemble framework of layered decision trees that beats deep neural networks on multiple datasets while resisting overfitting, even when training data is scarce. That robustness, the authors argue, makes it a strong candidate for real-world drug discovery pipelines where pristine datasets are a luxury rather than a given. These architectures matter because drug-target interaction prediction is the bottleneck that determines which molecules ever see the inside of a laboratory.

Generative AI is pushing the frontier even further. Generative adversarial networks and variational autoencoders, trained on massive chemical libraries like ZINC and ChEMBL, can design entirely novel molecules optimized for high binding affinity, low toxicity, and favorable ADMET profiles. Transformer-based models such as ChemBERTa apply natural language processing techniques to SMILES strings, the text-based representations of chemical structures, to predict properties like solubility and drug-likeness with competitive accuracy. Graph-based neural networks including GraphDTA and Mol2Vec encode molecular structures as graph embeddings, uncovering hidden bioactivity patterns in vast chemical spaces. And then there is AlphaFold 3, which delivers highly accurate predictions of protein, DNA, RNA, and ligand structures, effectively revolutionizing structure-based drug design and druggability assessments. Large language models are now layering on top of these tools, proposing new molecular entities with a speed and novelty that would have seemed like science fiction a decade ago.

The translational promise is no longer theoretical. Biopharmaceutical companies such as Exscientia and BenevolentAI have reported AI-discovered candidates advancing into Phase I and Phase II clinical trials, offering concrete proof that computational pipelines can produce molecules fit for human testing. AI is also proving adept at predicting synergistic drug combinations, with machine learning models trained on transcriptomic and proteomic data identifying novel drug pairs that show synergistic effects in preclinical cancer models. This is particularly valuable for stubborn malignancies like glioblastoma and pancreatic adenocarcinoma, where monotherapy routinely fails. Meanwhile, natural language processing tools such as PubTator mine the exploding biomedical literature to surface hidden associations between drugs, genes, and diseases, and their integration with large language models is supercharging hypothesis generation across the field.

Beyond the lab, AI is reshaping the full cancer care continuum. Deep learning algorithms now analyze mammograms, CT scans, and MRIs with radiologist-level accuracy, flagging suspicious images for urgent review and cutting false-negative rates in breast, lung, and prostate cancer screening. In pathology, whole-slide image classifiers have exceeded 90 percent accuracy in distinguishing benign from malignant lesions and in grading prostate and breast carcinomas, while also identifying predictive biomarkers like HER2 expression and PD-L1 status. One convolutional neural network trained on more than 40,000 CT scans matched or exceeded radiologists in detecting lung nodules. AI-driven auto-segmentation tools in radiotherapy delineate tumors and organs-at-risk with less inter-observer variability, and clinical decision support systems integrate genomic data with real-world evidence to recommend optimal treatment regimens. Wearable sensors paired with AI monitor recovering patients for complications like lymphedema, and models trained on electronic health records can flag patients at risk of sepsis or cardiotoxicity before symptoms become critical.

But raw accuracy, the review stresses, is not enough. The authors champion a concept they call model actionability, the degree to which an AI system can genuinely influence clinical or research decisions by reducing diagnostic and therapeutic uncertainty. One promising evaluation approach is decision impact analysis, which estimates how model recommendations change outcomes. In a prospective study of a machine learning tool predicting immunotherapy response in melanoma, the model’s guidance altered treatment strategies in 32 percent of patients and was associated with improved progression-free survival at six months. Interpretability techniques such as SHAP and LIME help reveal which features drive predictions, fostering clinician trust and easing regulatory review. Validation strategies including external cohorts and synthetic minority oversampling guard against overfitting, and the FDA’s proposed Total Product Life Cycle framework pushes sponsors to define predetermined change control plans so adaptive models can be updated post-approval without sacrificing safety oversight.

Regulators worldwide are scrambling to keep up. The FDA’s 2019 discussion paper on AI-based Software as a Medical Device outlined a lifecycle approach to oversight, while its Digital Health Software Precertification program evaluates companies rather than individual products, potentially streamlining approvals for AI-enabled oncology platforms. The European Medicines Agency has issued reflection papers emphasizing transparency, reproducibility, and explainability, warning that black-box models may face heightened scrutiny unless their outputs can be clearly justified. International bodies like the International Coalition of Medicines Regulatory Authorities are working toward harmonized principles, and sponsors now increasingly must submit validation datasets, explainability assessments, and performance benchmarks during investigational and new drug applications. Regulators are also demanding evidence that AI models generalize across populations, a critical concern in oncology where treatment responses vary by ethnicity, tumor subtype, and molecular profile.

Formidable challenges remain on the road to widespread adoption. Oncology data is often siloed, incomplete, and inconsistently formatted, and models trained on retrospective single-center datasets frequently fail to generalize. Algorithmic bias is a serious threat, with studies showing that some radiology and genomics models underperform in minority populations due to underrepresentation in training data. Federated learning offers a partial fix by training models on decentralized datasets without sharing raw patient information, preserving privacy while improving diversity. Interpretability, legal accountability for AI-driven errors, intellectual property questions around AI-generated discoveries, and workforce readiness all demand attention, and low-resource settings risk being left behind entirely. The authors’ prescription is a collaborative one: machine learning experts, oncologists, ethicists, regulators, and patient advocates working together toward transparent, equitable, and rigorously validated AI. If that vision holds, the future of cancer care may be more predictive, more personalized, and more patient-centered than anything the field has yet seen.

Subject of Research: Applications of artificial intelligence and machine learning in oncology drug development, clinical decision-making, and regulatory integration

Article Title: Advancing cancer care through artificial intelligence: from innovative models to clinical decision-making and regulatory integration

Article References: Subhan, A., & Manoharan, G. (2025). Advancing cancer care through artificial intelligence: from innovative models to clinical decision-making and regulatory integration. Clinical Cancer Bulletin, 4(1), Article 23. https://doi.org/10.1007/s44272-025-00052-0

Image Credits: AI Generated

DOI: 10.1007/s44272-025-00052-0

Keywords: artificial intelligence, oncology, drug discovery, machine learning, deep learning, AlphaFold 3, clinical decision support, FDA regulation, precision medicine, drug-target interaction, model actionability, federated learning

Sex pheromones are the linchpin of reproduction in most moths. Females synthesize and release species-specific blends of volatile molecules, usually during a defined window of the night, and males detect these blends over remarkable distances at astonishingly low concentrations. The precise composition of the blend, both its constituent molecules and their ratios, determines species identity and helps maintain reproductive isolation between closely related species that share habitat. Yet despite their central importance, pheromone compositions are well characterized almost exclusively for agricultural pest species, where the commercial incentive is clear: synthetic pheromones underpin population monitoring traps, mass trapping programs, and mating disruption strategies. Non-pest species, including the lily moth, whose caterpillars feed on ornamental Amaryllidaceae plants in gardens, have been largely overlooked.

Traditional pheromone identification is laborious and constrained. It typically requires access to living insects for gland extraction or headspace collection, careful determination of the females’ calling period, and painstaking analytical chemistry to pick out the behaviorally relevant compounds from complex extracts. The team sidestepped these hurdles by starting from a different vantage point altogether: the moth’s genome. Taking advantage of a recently assembled chromosome-level genome of S. picta, they searched for orthologs of OR5 and OR75, two pheromone receptors previously characterized in the sister species S. litura and S. littoralis, where OR5 is narrowly tuned to the major pheromone component (Z,E)-9,11-tetradecadienyl acetate, abbreviated (Z,E)-9,11-14:OAc.

The genomic search succeeded on both counts. The annotated S. picta genes, SpicOR5 and SpicOR75, encode proteins of 397 amino acids each, and quantitative PCR showed that SpicOR5 is strongly over-expressed relative to SpicOR75 in adult male antennae, a pattern mirroring the sister species and consistent with OR5’s established role as the principal receptor mediating male attraction in S. littoralis and S. litura. Expression levels were stable between one-day-old and three-day-old males, suggesting the receptors are not regulated by early adult sexual maturation. The sequence identities between S. picta receptors and their orthologs in the two sister species exceeded 94 percent, in sharp contrast to the roughly 67 to 72 percent identity they share with more distantly related Spodoptera species.

High sequence identity alone, however, is not a guarantee of conserved function. Research on other moths has shown that a single amino acid substitution can dramatically shift a pheromone receptor’s tuning. To address this uncertainty, the researchers built three-dimensional models of the receptors using the AlphaFold 3 web server and interrogated the physicochemical properties of their predicted ligand-binding pockets, including cavity volume and hydrophobicity, two parameters previously shown to correlate directly with odorant receptor tuning. The models of S. picta, S. litura, and S. littoralis receptors were nearly superimposable, with mean root-mean-square deviations of about 0.3 angstroms across the full-length proteins. Molecular docking then placed the candidate ligand (Z,E)-9,11-14:OAc in virtually the same binding pose within SpicOR5 as in its orthologs, engaging conserved residues across all three species.

Functional validation followed in a heterologous system. The S. picta receptors were expressed transgenically in Drosophila melanogaster olfactory neurons, replacing the fruit fly’s endogenous receptor in at1 sensilla, and tested against a panel of 26 moth pheromone compounds using single-sensillum recordings. As predicted, SpicOR5 responded to exactly one compound in the panel: (Z,E)-9,11-14:OAc, which elicited a median firing rate of 49 spikes per second and activated the receptor from doses as low as 10 micrograms. SpicOR75, true to its predicted broad tuning, responded significantly to seven of the 26 compounds, sharing most of its ligand repertoire with its orthologs in the sister species. The docking predictions and electrophysiology agreed almost perfectly.

With the receptor tuning established, the team flipped the logic of pheromone discovery. If SpicOR5, the dominant receptor in male antennae, is exquisitely tuned to (Z,E)-9,11-14:OAc, then this compound should be the major component of the female pheromone blend. Behavioral observations first pinpointed the optimal extraction window: female calling behavior peaked on the first two days after adult emergence, six and a half to nine hours into the dark phase, when up to 60 percent of females were calling. Gas chromatography coupled to electroantennographic detection of extracts from females collected during this window revealed a single compound that activated male antennae, and it was also the most prominent peak in the chromatogram. Subsequent mass spectrometry confirmed its identity as (Z,E)-9,11-14:OAc, alongside three minor components, (Z)-9-14:OAc, (E)-11-14:OAc, and (Z)-11-14:OAc, in a relative ratio of 100:45:34:3.

The behavioral assays sealed the conclusion. At 10 micrograms on filter paper, (Z,E)-9,11-14:OAc alone triggered wing-fanning in 56.3 percent of tested males, with additional males performing hair-pencil displays and contacting the odor source. A synthetic four-component blend matching the gland extract ratio elicited even more sexual behaviors, though still fewer than a natural female gland extract, which provoked the full behavioral sequence in every male tested. That shortfall echoes a pattern documented in the fall armyworm, S. frugiperda, where a trace pheromone component present at only about 0.2 percent of the major component’s abundance significantly enhances male attraction. The authors suggest that additional, still-unidentified trace compounds likely contribute to the complete S. picta pheromone signal.

Beyond its practical implications, the study illuminates a fascinating evolutionary story. OR75 exists only as a duplicated copy of OR5 in S. picta, S. litura, and S. littoralis; related species such as S. frugiperda and S. exigua, which do not produce (Z,E)-9,11-14:OAc, carry only a single, broadly tuned OR5 ortholog. The duplication apparently freed one copy to specialize. Multiple amino acid substitutions in the OR5 binding pocket shifted its specificity toward narrow recognition of the major pheromone component, while OR75 largely retained the ancestral broad tuning, though with modified specificity. Both copies thus represent cases of neofunctionalization relative to the ancestral receptor, and the functional divergence is mirrored by structural divergence in their binding sites, which cluster into two distinct groups that exclude the more distant Spodoptera orthologs.

The shared major pheromone component also raises ecological questions. S. picta and S. litura overlap geographically across large regions of South Asia and Oceania, and both call during the second half of the night, creating a potential risk of interspecific sexual signal interference. The two species’ minor components differ qualitatively, however, and these species-specific minor compounds may serve as critical discrimination cues for males, a mechanism documented in other Spodoptera species where heterospecific pheromone components act as behavioral antagonists. Host plant volatiles may further refine mate finding, since S. picta specializes on certain Amaryllidaceae plants, while its polyphagous sister species does not.

For applied entomology, the payoff is immediate. Identifying (Z,E)-9,11-14:OAc as a sufficient male attractant in S. picta opens the door to pheromone-based monitoring and trapping in gardens and nurseries where synthetic pesticides are restricted. More broadly, the study demonstrates a generalizable workflow: mine a genome for candidate pheromone receptors, model their structures, dock candidate ligands, verify with electrophysiology, and only then commit to chemical analysis. As genomic resources proliferate and AI-driven structure prediction matures, this receptor-guided strategy promises to bring chemical ecology within reach for the vast majority of insect species that will never enjoy the attention devoted to a handful of crop pests.

Subject of Research: Structure-based prediction of moth sex pheromone components from odorant receptor models in Spodoptera picta

Article Title: Odorant receptor structures predict the major female sex pheromone component in a moth

Article References: Comte, A., Wei, Z., Li, H., Xing, W.-Z., Moracci, R., Zhang, J., Fiorucci, S., & Jacquin-Joly, E. (2026). Odorant receptor structures predict the major female sex pheromone component in a moth. BMC Biology, 24(1), Article 193. https://doi.org/10.1186/s12915-026-02717-1

Image Credits: AI Generated

DOI: 10.1186/s12915-026-02717-1

Keywords: Spodoptera picta, sex pheromone, pheromone receptors, odorant receptors, AlphaFold 3, molecular docking, reverse chemical ecology, GC-MS, electrophysiology, gene duplication, reproductive isolation, pest management