The research team, led by Andrew Disharoon of the Medical University of South Carolina and Aaron J. Masino of Clemson University, assembled a dataset of 377,628 drug-pair samples drawn from DrugBank and the RxNav Drug and Drug Interaction dataset, with chemical structures retrieved through the PubChem API and therapeutic target information from the Therapeutic Target Database. Rather than relying on a small set of broad interaction categories, the team processed the free-text interaction descriptions in these databases—for example, sentences like “the metabolism of (R)-warfarin can be increased when combined with acetaminophen”—and extracted 168 normalized interaction templates by masking out the drug names. Each sample therefore consists of a drug pair, structural and target information for both compounds, and a set of binary labels indicating which of the 168 interaction types apply. Careful preprocessing removed formulation-specific terms such as “hydrochloride” or “maleate” so that, for instance, metoprolol succinate and metoprolol tartrate are treated as the same base drug, and a synonym list unified names like acetaminophen and paracetamol. Interactions with fewer than 50 samples were dropped to ensure enough data for reliable training.

A central design choice distinguishes InfoMedex from earlier approaches: the way molecules are turned into sequences a neural network can read. Instead of treating SMILES strings—compact text encodings of molecular structure—as ordinary character strings, the team converted them to Atom-in-SMILES notation, in which each token explicitly encodes an atom together with its local bonding environment, including ring membership, aromaticity, and attached hydrogens. This chemically meaningful tokenization preserves structural context that character-level splitting or compressed molecular fingerprints can lose, and it is particularly well suited to transformer architectures, where positional embeddings and self-attention can capture relationships between distant substructures. Prior work has shown that Atom-in-SMILES representations better capture local chemical context than character-level tokenization, facilitating substructure-level reasoning.

The architecture itself is a hybrid of transformer and convolutional components processing three input streams. Tokenized Atom-in-SMILES sequences for the drug of interest and for the interacting drug are each converted into dense embeddings of dimension 512 with learned positional encodings. Gene features derived from the Therapeutic Target Database, one-hot encoded across 1,143 features, pass through a two-layer feed-forward network into the same embedding space and are replicated across the sequence length. The three streams are concatenated into a unified sequence of size 3,072 and processed by a transformer encoder with eight attention heads, pre-layer normalization, and padding masks to keep padding tokens out of attention computations. A convolutional stack of four 1D layers with 124 filters and kernel size 7, ReLU activations, a dropout layer, and a linear classification head then produces logits for all 168 interaction labels. The full model contains roughly 70.35 million learnable parameters and was trained with sigmoid focal loss, an objective designed for severe class imbalance, reaching peak validation performance after an average of 22.8 epochs.

The performance results are striking, though they come with an important caveat. Under the primary evaluation, in which drug pairs are split randomly but individual drugs may appear in both training and test sets, the model achieved a micro-averaged area under the precision-recall curve of 0.857 and a macro mean AUPRC of 0.825 across the 168 labels. Accuracy averaged 0.997 and AUC-ROC 0.998, while precision-recall analysis showed the model maintaining precision above 70 percent at recall values as high as 80 percent. Given that only 0.603 percent of the more than 6.3 million test predictions were positive labels, the researchers deliberately relied on precision-recall curves rather than ROC curves, which prior studies have shown to be essentially uninformative under extreme class imbalance. Stratified analysis across frequency quantiles—from interactions positive in as few as 0.013 percent of samples up to the most common at 13.4 percent—showed median precision above 0.8 across all quantiles. For three very rare interaction types the model’s predicted probabilities failed to exceed the 0.5 decision threshold, but the probability distributions for positive and negative samples were largely separated, suggesting appropriate model uncertainty rather than outright failure.

That caveat emerged from a secondary, more demanding evaluation. When the team held out 10 percent of the drugs—86 compounds appearing as the drug of interest in at least one sample—and excluded every sample containing them, mean AUPRC fell from 0.825 to 0.341 across 142 evaluable labels. In other words, InfoMedex generalizes well to unseen drug pairs involving drugs it has encountered elsewhere in training, but predicting interactions for entirely unseen molecules remains substantially harder. This distinction matters enormously for the model’s ultimate use case: flagging risks for newly approved drugs that have no interaction history at all. The authors are candid that this limitation should temper expectations for near-term deployment on novel compounds.

Perhaps the most intriguing contribution is the explainability analysis. Using Integrated Gradients, a gradient-based attribution method implemented in the Captum library, the team computed the contribution of each atom token to selected predictions, aggregating per-sample ranks across many partner drugs to obtain a robust estimate of which molecular regions matter most. Because a given drug’s token sequence and input position are fixed across samples, the averaging marginalizes only over the partner drug, yielding a well-defined estimand: the typical within-drug importance ranking of each token over the empirical distribution of partners. Rank aggregation was chosen over magnitude-based averaging because it is robust to samples carrying disproportionate attribution mass, and interquartile ranges were reported to measure rank stability.

The case studies focused on two well-characterized cytochrome P450 inhibitors. For bupropion, a CYP2D6 inhibitor represented by 29 atom tokens, the highest-importance tokens clustered on the carbonyl-centered motif—the ketone carbonyl carbon had a mean rank of 1.88—and on the bulky N-alkyl tert-butylamine region, including the quaternary carbon at mean rank 6.42 and the basic nitrogen. Both patterns align with known pharmacology: the carbonyl is the starting point for reductive metabolites that show stronger CYP2D6 inhibition than the parent drug, and CYP2D6 substrates and inhibitors frequently present a basic nitrogen that contributes to binding in the enzyme’s tight active site. The meta-chloro phenyl region also ranked prominently, consistent with evidence that this motif discourages aromatic hydroxylation and redirects metabolism toward CYP2B6, which produces hydroxybupropion—the metabolite responsible for roughly two-thirds of bupropion’s in vivo CYP2D6-inhibiting activity.

Ritonavir, the HIV protease inhibitor deployed clinically precisely because of its potent CYP3A4 inhibition, presented a different picture. Represented by 83 atom tokens with 18 points of rotation, the molecule drew the model’s strongest emphasis to the central scaffold, where the hydroxyl token had the highest mean rank of 1.39 with a narrow interquartile range of 1.0 to 1.75 across 18 partner drugs, followed by an adjacent stereocenter. These atoms sit near the benzyl and phenyl bulk that provides steric stabilization, helping hold the drug in a binding-ready configuration within the CYP3A4 cavity. Structural studies have shown ritonavir is a type II ligand that coordinates the heme iron through a thiazole nitrogen while packing hydrophobic groups tightly into the active site, effectively locking the molecule into a high-affinity pose with a slow off-rate. The model also emphasized carbonyl-linked oxygen tokens and the methylene linker connecting the heteroaryl ring to the tertiary amine arm, features that plausibly govern how the bulky aromatic half and the heme-facing heteroaryl half engage the enzyme. Notably, some highly ranked tokens were non-atom structural signposts such as parentheses, which the authors suggest the model uses to guide its structural reading of the molecule rather than as chemically interpretable features.

Taken together, the results position InfoMedex as a scalable, interpretable complement to post-marketing surveillance and formal interaction studies, of which only a limited subset of clinically relevant combinations can feasibly be evaluated under FDA guidance. The ablation finding that randomly perturbing the target features did not significantly change performance metrics suggests the current one-hot gene encodings underexploit mechanistic context, and the authors propose graph neural networks or cross-attention architectures as avenues to better fuse drug-target data. The held-out-drug results temper enthusiasm for novel compounds, and the explainability analysis covered only interactions with well-studied CYP-mediated mechanisms. Still, the demonstration that a transformer reading molecules atom by atom can both predict 168 interaction types with high precision and point to the specific chemical motifs responsible marks a meaningful step toward AI systems that do not just flag drug interactions but can generate mechanistic hypotheses about why they occur—exactly the kind of transparent reasoning that clinical and regulatory adoption will demand.

Subject of Research: Deep learning prediction and explainability of drug-drug interactions using Atom-in-SMILES tokenization and a multimodal CNN-transformer architecture

Article Title: InfoMedex: drug-drug interaction prediction via a multimodal CNN-transformer model

Article References: Disharoon, A., Pasupuleti, S., Thurston, C., Fiore, M., McKenzie, S., Smith, A., & Masino, A. J. (2026). InfoMedex: drug-drug interaction prediction via a multimodal CNN-transformer model. Frontiers in Drug Safety and Regulation, 6, Article 1841035. https://doi.org/10.3389/fdsfr.2026.1841035

Image Credits: AI Generated

DOI: 10.3389/fdsfr.2026.1841035

Keywords: drug-drug interactions, deep learning, transformer, Atom-in-SMILES, explainable AI, Integrated Gradients, cytochrome P450, pharmacovigilance, DrugBank, bupropion, ritonavir, precision medicine

The research, led by Sarah Asbury and Jane A. Foster of McMaster University together with Madhukar H. Trivedi of the University of Texas Southwestern Medical Center and colleagues, rests on a deceptively simple insight. Although the composition of a person’s gut microbiome is shaped by diet, environment, and countless life experiences, individual bacterial species are not entirely random residents. Twin studies and large population analyses have shown that the abundance of many gut microbes has a heritable component, meaning that human genetic variation influences which bacteria thrive in the intestine. If host genes help determine microbial makeup, then a person’s genome should carry statistical fingerprints of their microbiome, and those fingerprints might be quantified the same way geneticists already quantify inherited risk for diseases.

That quantification tool is the polygenic risk score, or PRS, a method that has become a workhorse of modern genetics. A PRS is built by aggregating the effects of thousands or millions of genetic variants across the genome, each weighted by how strongly it has been associated with a trait in a large discovery sample. Instead of applying this technique to a human disease, the team applied it to bacteria. Using data from the Texas Resilience Against Depression study, known as T-RAD, they computed polygenic risk scores for 25 heritable gut bacterial taxa, effectively generating a genome-based estimate of each participant’s propensity to host particular microbes. The approach is novel in neuropsychiatric research, where the gut-brain axis has attracted intense interest but has lacked genetically anchored biomarkers.

The gut-brain axis, the bidirectional communication network linking the intestinal microbiota with the central nervous system, has been implicated in depression through multiple mechanisms. Gut microbes produce and modify neurotransmitters, generate metabolites that cross the blood-brain barrier, modulate immune signaling, and influence the stress-response system. Animal studies have shown that transferring gut bacteria from depressed rodents or humans into germ-free animals can transmit depressive-like behaviors. What has been missing is a way to translate this biology into a clinically useful predictor, one that does not require repeatedly sampling a patient’s stool but can instead be read out from a blood or saliva sample used for standard genotyping.

To test whether microbial PRS could fill that gap, the researchers turned to EMBARC, the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care study, one of the most rigorously designed biomarker studies in depression research. EMBARC enrolled 226 participants with depression across four academic medical centers: UT Southwestern in Dallas, Massachusetts General Hospital in Boston, the University of Michigan in Ann Arbor, and Columbia University in New York. Participants were treated with sertraline, a widely prescribed selective serotonin reuptake inhibitor, and followed over time with careful measurement of depression severity. The design allowed the team to compute the same microbiome-related genetic scores in EMBARC participants and ask whether the scores identified in T-RAD held predictive value in an independent cohort, a validation strategy that lends considerably more weight to the findings than a single-cohort analysis would.

The results were striking in their specificity. In logistic regression analyses, three of the microbiome-related polygenic risk scores showed significant predictive value for remission after eight weeks of sertraline treatment, the standard early checkpoint in antidepressant trials. A fourth score predicted remission at sixteen weeks, following augmentation of sertraline with bupropion, a different mechanism of action often added when the first medication falls short. Beyond remission, four of the microbial scores were associated with greater changes in depression severity over time in patients receiving sertraline compared with placebo, suggesting that the genetic microbial signatures tracked not just a binary outcome but the trajectory of symptoms themselves.

For the authors, the significance lies less in any single bacterium than in the demonstration of principle. By applying established polygenic risk score methodology to the microbiome, the study shows that it is possible to leverage the gut microbiome to identify microbiome-related genetic predictors of antidepressant response without directly measuring the microbes at all. The scores function as a bridge between two data types that are rarely available together: detailed microbiome sequencing, which is expensive and technically demanding, and routine human genotyping, which is increasingly cheap and scalable. If the approach matures, a clinician could one day order a genetic test that implicitly encodes microbial information and use it to guide the initial choice of antidepressant, potentially sparing patients months of ineffective treatment.

The work also carries broader implications for how biomarkers are developed in psychiatry. The field has long struggled to find biological measures that predict treatment response, with candidates ranging from brain imaging patterns to blood-based gene expression panels. Most have failed to replicate across cohorts, a problem the present study explicitly tried to avoid by validating its scores in the independent EMBARC sample. The fact that the predictive signals were specific to particular drugs and timepoints, rather than generic markers of recovery, suggests the scores capture something biologically meaningful about how an individual’s gut-brain axis interacts with serotonergic and noradrenergic medications. That specificity, if it holds up, is exactly what precision psychiatry requires.

Caution is nonetheless warranted before the findings reach the clinic. The EMBARC validation cohort comprised 226 participants, a modest number by the standards of polygenic prediction, where effect sizes are typically small and large samples are needed to establish robust performance. The discovery work in T-RAD and the validation in EMBARC both involved participants recruited within specialized research programs, and it remains to be seen whether the same scores predict response in more heterogeneous primary care populations, in patients taking other antidepressants, or across different ancestries, since polygenic scores are known to lose accuracy when applied to populations different from those in which they were derived. The study was funded by the Center for Depression and Clinical Research at UT Southwestern and the Ontario Brain Institute, and both underlying studies were registered on clinicaltrials.gov, with T-RAD under NCT02919280 and EMBARC under NCT01407094, providing a transparent foundation for the replication efforts that must now follow.

Even with those caveats, the study opens a genuinely new frontier at the intersection of genomics, microbiome science, and treatment prediction. It suggests that the heritable architecture of the gut microbiome is not merely a curiosity but a usable signal, one that connects the biology of the gut-brain axis to the practical question every depressed patient and clinician faces: which treatment is most likely to work, and how soon. As the authors note, the result points toward a promising new approach to biomarker development in depression. If larger studies confirm and refine these microbial genetic predictors, the humble gut bacterium may earn an unlikely role in one of medicine’s most consequential decisions, helping to end the era of trial-and-error prescribing for millions of people worldwide.

Subject of Research: Using microbiome-related polygenic risk scores to predict antidepressant treatment response in depression

Article Title: Microbiome-related polygenic risk scores and treatment response in depression

Article References: Asbury, S., Liaqat, H., Chin Fatt, C., Seelam, N., Elsheikh, S. S. M., Müller, D. J., Vakorin, V. A., Moreno, S., Trivedi, M. H., & Foster, J. A. (2026). Microbiome-related polygenic risk scores and treatment response in depression. Translational Psychiatry. https://doi.org/10.1038/s41398-026-04472-x

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04472-x

Keywords: depression, gut microbiome, polygenic risk score, antidepressant response, sertraline, bupropion, gut-brain axis, biomarkers, EMBARC, precision psychiatry, Translational Psychiatry, treatment prediction

The clinical stakes are enormous. The review opens with the striking evidence from the United Kingdom Million Women Study, which quantified how stopping smoking at roughly 30, 40, or 50 years of age recovers years of life that would otherwise be lost. Against that backdrop, the authors survey the established pharmacopeia of cessation: bupropion, varenicline, nortriptyline, and clonidine. Each acts on a different node of nicotine’s addictive circuitry. Bupropion, an aminoketone antidepressant, inhibits the dopamine and norepinephrine transporters and antagonizes nicotinic acetylcholine receptors, blunting both withdrawal symptoms and relapse risk. Varenicline, a partial agonist at the α4β2 nicotinic receptor, partially mimics the subjective reward of smoking while blocking nicotine’s full activation of the receptor, delivering about 45 percent of nicotine’s maximal effect and attenuating dopamine release induced by the drug itself.

The clinical trial record assembled in the review is remarkable for its breadth. A randomized trial of nortriptyline combined with behavioral counseling achieved a six-month cessation rate of 14 percent versus 3 percent for placebo, with significant reductions in withdrawal symptoms such as anxiety, irritability, and difficulty concentrating. Adding transdermal nicotine to nortriptyline pushed six-month abstinence to 23 percent versus 10 percent with placebo. On the varenicline front, a Chinese randomized controlled trial in smokers with chronic obstructive pulmonary disease reported continuous abstinence of 43.1 percent with varenicline versus 23.5 percent with bupropion among normal nicotine metabolizers. Even vaping cessation is now in scope: a double-blind placebo-controlled trial found biochemically validated continuous abstinence of 40 percent with varenicline plus counseling versus 20 percent with placebo, suggesting the drug’s reach extends beyond traditional cigarettes.

Perhaps the most clinically provocative thread is the nicotine metabolite ratio, or NMR, the ratio of 3′-hydroxycotinine to cotinine that reflects how quickly an individual clears nicotine. In a landmark randomized trial of 1,246 participants, varenicline outperformed the nicotine patch in normal metabolizers, while slow metabolizers benefited equally from the cheaper patch and suffered more varenicline side effects. A 2024 trial in smokers with COPD sharpened the picture further: slow metabolizers experienced more adverse effects from varenicline than bupropion, and no efficacy gap between the two drugs. These findings move the field toward genetically informed, biomarker-guided prescribing, in which a simple metabolic measurement could steer a smoker toward the therapy most likely to work with the fewest harms.

Bupropion’s analytical dossier illustrates the sophistication of modern pharmaceutical quality science. Yeniceli and Dogrukol-Ak developed a thin-layer chromatography method on silica gel plates using an ethanol-chloroform-glacial acetic acid mobile phase, achieving linearity from 200 to 1000 nanograms per band with detection limits near 11 nanograms and precision below 2 percent relative standard deviation. Borges and colleagues built a high-throughput LC-MS/MS assay on a monolithic column that separated bupropion and its metabolites hydroxybupropion and threo-hydrobupropion from human, mouse, and rat plasma in as little as 23 seconds, with isotope-labeled internal standards ensuring sensitivity down to a quarter of a nanogram per milliliter. Meanwhile, a Design of Experiments-driven enantioseparation by HPTLC resolved the enantiomers of bupropion and its active metabolite with resolution factors above 6, underscoring why chirality matters when the two mirror-image forms of a drug can behave differently in the body.

The review also highlights the rise of Analytical Quality by Design, or AQbD, a paradigm that replaces trial-and-error method development with predefined analytical targets, risk assessment, and statistically designed optimization. A striking example is the chaotropic chromatography method developed for bupropion and its five impurities: a Box-Behnken design explored critical method parameters, Monte Carlo simulations defined a design space with at least an 85 percent probability of meeting acceptance criteria, and the validated method was applied directly to commercial Wellbutrin tablets. Similar rigor appears in varenicline analysis, where stability-indicating HPLC methods resolved the drug from forced-degradation products, and where researchers isolated and structurally characterized a previously unknown impurity, 4,6,7,8,9,10-hexahydro-1H-6,10-methanopyrazino[2,3-h]benzazepine-2,3-dione, present at 0.2 percent in tablet samples.

Nortriptyline research adds a delivery-science dimension. Several groups have engineered transdermal patches using hydroxypropyl-methyl-cellulose or chitosan matrices, tuning permeation enhancers such as propylene glycol, ethanol, oleic acid, and polysorbate 80 to push the drug across human skin. Flux values spanning roughly 20 to 256 micrograms per square centimeter per hour mean a patch only 2 to 3.5 centimeters wide could deliver the 25 to 75 milligrams needed daily for cessation therapy, and confocal microscopy confirmed no skin damage. In silico-in vitro extrapolation went further: a one-compartment transport model built from infinite-dose permeation experiments predicted in vivo plasma levels in rats within the therapeutic window, above the 40 nanograms per milliliter threshold associated with cessation benefit. Electrochemical methods, including cathodic adsorptive stripping voltammetry, round out the analytical toolkit with detection limits around 50 nanograms per milliliter.

Clonidine brings a different pharmacological lesson: response is not uniform. The classic 1988 double-blind trial by Glassman and colleagues found clonidine-treated heavy smokers achieved confirmed abstinence at more than twice the rate of placebo, verified by serum cotinine, but the effect was far stronger in women than in men, and a history of major depression predicted relapse regardless of treatment. The drug stimulates presynaptic alpha-2 adrenergic receptors in the brainstem, damping sympathetic outflow and easing irritability, anxiety, and craving. Analytically, clonidine is now measurable by paper spray tandem mass spectrometry, a column-free technique as accurate as conventional HPLC-MS/MS, and by sensitive LC-MS/MS assays in plasma that supported bioequivalence studies of 25-microgram tablets. Forced degradation studies show clonidine is robust under acidic and neutral stress but vulnerable to oxidation, a finding that directly informs storage and formulation decisions.

What unifies this sprawling body of work is the review’s insistence that method validation and sustainability are no longer optional extras. Every method discussed is judged on specificity, accuracy, precision, linearity, range, and robustness under ICH Q2(R2) guidance, and increasingly through green chemistry metrics. The Analytical Eco-Scale scores procedures on reagent hazard, energy use, and waste; AGREE assigns a holistic sustainability score; MoGAPI maps environmental impact across a method’s lifecycle; and the emerging White Analytical Chemistry framework adds red, or analytical-performance, and blue, or practical-applicability, dimensions to the green calculus. A Multi-Color Assessment spanning all four dimensions, the authors argue, is becoming the benchmark for methods fit for routine quality control, pharmacokinetic studies, and bioequivalence testing alike.

The big picture emerging from this synthesis is that quitting smoking is being transformed by data on both sides of the prescription pad. Biomarkers like the nicotine metabolite ratio promise to match patients to varenicline, bupropion, nortriptyline, or patches before the first dose is taken, while validated, stability-indicating, and increasingly green analytical methods guarantee that every tablet delivers exactly what the label claims, even years into its shelf life. The authors point toward combination therapies, innovations in transdermal delivery, gender-specific treatment responses, and high-throughput hybrid analytical platforms as the next frontier. If the clinical and analytical threads continue to braid together at this pace, the humble act of putting out a cigarette for good may soon rest on a foundation of biomarkers, design-of-experiments chromatograms, and sustainability scores, a distinctly twenty-first-century recipe for attacking one of humanity’s oldest addictions.

Subject of Research: Clinical, pharmacological, and analytical review of smoking cessation pharmaceutical drugs

Article Title: A review of clinical, pharmacological and analytical aspects of smoking cessation pharmaceutical drugs

Article References: Bagul, V. A., & Ambadekar, S. R. (2026). A review of clinical, pharmacological and analytical aspects of smoking cessation pharmaceutical drugs. Discover Industrial Chemistry and Materials, 1(1), Article 18. https://doi.org/10.1007/s44508-026-00019-6

Image Credits: AI Generated

DOI: 10.1007/s44508-026-00019-6

Keywords: smoking cessation, bupropion, varenicline, nortriptyline, clonidine, HPLC, LC-MS/MS, nicotine metabolite ratio, forced degradation, impurity profiling, Analytical Quality by Design, green analytical chemistry