Anurendra Kumar

Bio

Postdoctoral Researcher, Stanford University: computational genomics, disease modeling, AI for drug discovery.

From association to mechanism: disease programs, spatial genomics, and perturbation data for drug prioritization.
Modern biomedicine can now measure molecular state at unprecedented scale, but turning these measurements into mechanistic insight and reliable therapies remains difficult. In omics data, a major reason is structured confounding: tissue architecture, local composition, and spatial context can produce convincing patterns that fail to reproduce or translate. My research develops statistical and machine learning frameworks to account for this structure and to enable biologically grounded disease modeling and drug prioritization.

I am driven by a simple question: how can we use complex genomic data to understand why disease unfolds the way it does, and how can that understanding lead to better treatment? My current work focuses on modeling disease programs from single-cell and spatial omics data and linking them to therapeutic interventions. I integrate scRNA-seq, spatial transcriptomics, perturbation datasets, and causal modeling to identify mechanistically grounded targets in complex disease, with particular emphasis on immune evasion in lung cancer and genetic cardiac diseases such as dilated cardiomyopathy (DCM). A central goal is to distinguish disease-specific signals from shared remodeling programs and to identify interventions that selectively reverse the disease-relevant components.

A growing focus is transportable perturbation modeling: understanding what is conserved versus context-specific in intervention responses, and when signals from large perturbation atlases can be transferred to disease-relevant systems such as iPSC-derived or organoid models. This creates a closed loop from computational prioritization to experimental validation through collaborations, enabling more reliable translation of omics-derived hypotheses.

The through-line across this work is causality: without it, even rich spatial or perturbational data often produces prioritization lists that do not translate. My PhD work on subcellular RNA organization (InSTAnT, Nature Communications 2024) and cell–cell communication in spatial data (CellWHISPER, bioRxiv 2026) was driven by the same insight: structured confounding is a dominant failure mode in spatial biology, and conditioning on it can change biological conclusions.

I have also interned at Genentech, Vertex, IBM Research (4×), and others, and briefly explored a startup, experiences that taught me a great deal about building end-to-end systems.