Mech(AI)nistic (BioRxiv; Manuscript in review)
Genome-scale metabolic modeling
is an incredibly useful tool to understand how diseases work on at a cellular scale, but the technical
barrier to entry can be quite high because it requires knowledge in programming and metabolic/mechanistic
modeling. To overcome this gap, I, leading several undergraduate students, developed
MechAInistic. It’s a multi-agent system that uses Large Language Models to turn natural
language questions into executable, model-grounded workflows. By using an "Architect-Reviewer" design, the
system builds its own analysis pipelines and returns structured reports on everything from drug-target
exploration to pathway comparisons. To validate MechAInistic, we tested it against to identify a drug
therapy for (1) Naive B cells in a Rheumatoid Arthritis context, and (2) CD4+ Th17 cells in a Multiple
Sclerosis context. In the Naive B cells, MechAInistic nominated Devimistat, and for the Th17 cells, it
recommended Ivosidenib.
Constraint-based Optimization of Metabolic Objectives (COMO) is a comprehensive, user-friendly pipeline
designed to streamline the integration of multi-omics data for genome-scale metabolic modeling (GSMN) and
drug discovery. By combining heterogeneous datasets (including bulk and single-cell RNA-seq, microarrays,
and proteomics) with GSMNs, COMO allows researchers to efficiently construct context-specific models for
various cell and tissue types within a unified Docker or Conda environment. The pipeline automates complex
tasks such as data processing, simulation, and drug perturbation analysis to identify potential therapeutic
targets and repurposable drugs. COMO was validated through a study on B cell metabolism in rheumatoid
arthritis and systemic lupus erythematosus and offers a robust computational solution for accelerating the
discovery of low-cost, effective disease treatments.
AutoRNAseq is a highly parallelized Snakemake workflow designed to automate the processing of bulk RNA-seq
data on high-performance computing (HPC) clusters. Tailored to interface with NCBI's
Gene Expression Omnibus
Database, the pipeline requires only a list of SRR codes as input. It handles parallel downloading and
unpacking of raw data, performs quality control (using
FastQC
and
MultiQC), optionally trims reads (Trim Galore!), aligns sequences to the genome using
STAR
and performs gene quantification with
Salmon. While specifically optimized to format inputs for metabolic drug discovery and repurposing packages
(specifically COMO), AutoRNAseq serves as a robust, standalone solution for any
individual needing to efficiently generate standardized gene count matrices and quality metrics for
downstream RNA-seq analysis.
