Sriram Chandrasekaran

Machine learning and genome-scale metabolic modeling have accelerated the discovery of metabolic-regulatory relationships in model microbes and mammalian cells. This review provides a roadmap for integrating omics with computational methods for studying systems-level transcriptional, post-translational, and allosteric regulation. Developing methods to infer and integrate multiple regulatory mechanisms simultaneously is essential for discovering novel modes of metabolic regulation and precise engineering of metabolic networks.

To explore multiomic regulation of the metabolome, we used machine learning to predict metabolomic variation across ~1000 different cancer cell lines with matched omics data from eight biomolecular classes: genomic copy number variation, mutations, DNA methylation, histone post-translational modifications (PTMs), transcriptomics and RNA splice variants, non-coding transcriptomics (miRNA and lncRNA), proteomics, and phosphoproteomics. Overall, the metabolome is tightly associated with the transcriptome, with coding and non-coding RNAs emerging as top predictors. Peripheral metabolites are predictable via levels of corresponding enzymes, while those in central metabolism require combinatorial predictors in signaling and redox pathways, and may not reflect corresponding pathway expression. We reconstruct multiomic interaction subnetworks for highly predictable metabolites, and YAP1 signaling emerged as a top global predictor across four omic layers. We prioritize predictive multiomic features for single-cell and spatial metabolomics assays. Top predictors were enriched for synthetic-lethal interactions and synergistic combination therapies that target compensatory metabolic modulators.

With nearly 90% of drug candidates failing in clinical trials due to poor efficacy or toxicity, drug repurposing has emerged as a vital strategy to accelerate the delivery of life-saving treatments. However, most current drug repurposing approaches fail to account for the physiological barriers and downstream biological impacts that dictate therapeutic success. To bridge this gap, we present SCOUT (Screening Candidates via Organ Uptake and Target-binding), a modular machine learning-driven framework for drug repurposing by simultaneously modeling organ permeability, drug-target binding, and biomarker modulation. Unlike conventional repurposing efforts that rely on single-point predictions or disconnected steps, SCOUT integrates these factors into a unified predictive funnel, significantly reducing trial-and-error and prioritizing candidates with the highest probability of therapeutic relevance. As a proof of concept, we applied SCOUT to Alzheimer’s disease, where blood-brain barrier (BBB) penetration is an important hurdle. The framework first predicts the unbound brain-to-plasma partition coefficient, Kpuu, using SMILES-based embeddings and achieved robust performance (Accuracy: 0.90, Recall: 0.94), then integrates machine learning to predict binding affinity with BACE1, a key Alzheimer’s target that was used in this work as a proof of concept. By combining these modules, SCOUT reduced the candidate search space by 99.9% and identified hits with diverse mechanisms of action. Critically, SCOUT extends beyond hit identification by employing mechanistic modeling to simulate how these candidates might modulate specific disease biomarkers (e.g. Amyloid-beta (Aβ) peptides). This hybrid approach ensures repurposed candidates are not only chemically viable but also physiologically active, providing a rational and resource-efficient pipeline for drug development.

Antimicrobial resistance poses a major global threat, driven by diminishing efficacy of current treatments and limited new therapies. Combination therapy with existing drugs offers a promising solution, yet current empirical screening methods are expensive and often lead to suboptimal efficacy and inadvertent toxicity. We introduce CALMA, a computational framework that quantitatively analyzes the potency-toxicity landscape of multi-drug combinations. Integrating genome-scale metabolic modeling with a neural network that reflects metabolic subsystems, CALMA enhances interpretability and prioritizes pathways influencing drug interactions. The incorporation of metabolic architecture in the neural network leads to over 92% reduction in model parameters, enabling it to learn generalizable mechanistic signals and reducing the experimental search space of optimal combinations by 97%. CALMA identified promising antimicrobial combinations against Escherichia coli and Mycobacterium tuberculosis that were antagonistic for kidney and liver toxicity and uncovered the nucleotide salvage pathway as a selective influencer of toxicity, which was validated in vitro. Mining of health records of over 400,000 patients showed reduced frequency of kidney side-effects in patients taking a vancomycin combination identified by CALMA. CALMA provides a rational, mechanistic approach to streamline combination treatment design.

Differences in microbiome composition profoundly influence drug response, yet methods to model the metabolic interplay between tumors, microbes, and therapeutics remain limited. We present a generalizable framework combining machine-learning and genome-scale metabolic modeling to prioritize combination therapies for colorectal cancer (CRC) in the presence of Fusobacterium nucleatum ( Fn ) and other pathogenic, probiotic, and commensal microbes. Trained on 6,514 drug combinations in microbe-free CRC cell lines, the model predicted synergistic combinations in both microbe-free and microbe-associated contexts and generalized to immunotherapy-associated conditions. Predictions were validated using an asymmetric co-culture system that mimics the colon’s normoxic–anaerobic gradient, confirming synergistic combinations in HCT116 cells with Fn , including drugs not typically used in CRC therapy. Mechanistic analysis and targeted pharmacological perturbations revealed phospho-inositol metabolism and cysteine transport as key determinants of Fn -dependent drug synergy. Together, this work introduces a scalable, microbiome-aware framework to enable discovery of context-specific combination therapies.