Anurendra Kumar

Postdoctoral Scholar, Stanford Cardiovascular Institute (CVI)

anurendk@stanford.edu

Mechanism-aware AI for genomics.

Models in genomics break when you move them. I work on why.

I am a postdoc with Joseph C. Wu, co-mentored by Luca Pinello and supported by the Sandy and Joan Weill CVI Postdoctoral Grant. I did my PhD with Saurabh Sinha at Georgia Tech.

I want to understand why disease unfolds the way it does, well enough to treat it. Two kinds of data speak to this, and they sit at opposite ends of the same causal chain.

A perturbation screen gives a cause and an effect days apart. High resolution, no outcome at the organism level.

Human genetics gives a cause and an effect four decades apart. A real intervention, a real endpoint, nothing in between.

The middle is missing. Mechanism is what fills it.

I work at the level of cells and genes, the finest resolution at which that chain is still visible, and specifically on signaling. Move a model to a new cell type, a new patient, or a new perturbation, and performance quietly collapses. My working hypothesis is that mechanism is what buys generalization. In modeling terms it is an inductive bias, and one of the few with a plausible claim to being true. Statistics is the other half of the method. It tells me when the data cannot support the structure I want to impose. Over the next twenty years, I want a drug for cardiovascular disease to come out the other end.

Schematic: disease and perturbation atlases feed a mechanism-aware model across a context shift; the model prioritizes interventions that are tested in iPSC-derived cardiomyocytes, which feed back to falsify and refine the model.

My focus is genetic disease, starting with dilated cardiomyopathy. Nearly all large-scale perturbation data comes from cancer cell lines, because they are cheap to screen. The cells that matter are cardiomyocytes. What carries across that gap, and what does not, is the question I keep coming back to.

Most of my work so far has been in spatial omics, one of the few places the middle is directly visible: cells in real tissue, and which of them talk to each other. The question running through it is how a ligand's effect propagates through cells: who sends it, who receives it, and what changes downstream. CellWHISPER (bioRxiv 2026; code) is the first published piece of it, and won second prize at the 2025 Eric and Wendy Schmidt Center Symposium at MIT. The same idea now shapes my current projects. The statistical half started in my PhD, with the finding that spatial data is dominated by structured confounding: tissue architecture and local composition manufacture convincing patterns (InSTAnT, Nature Communications 2024; code, talk).

Previously at Genentech, Vertex, and IBM Research, with a brief detour into building a startup, where I learned what it takes to make an end-to-end system actually work.

Publications

Most recent publications on Google Scholar.
‡ indicates equal contribution.

CellWHISPER disentangles direct cell-cell communication from structural proximity

Anurendra Kumar, Felix Rivera Moctezuma, Bhavay Aggarwal, Nicholas Zhang, Ahmet F Coskun & Saurabh Sinha

Biorxiv 2026

Code

Intracellular spatial transcriptomic analysis toolkit (InSTAnT)

Anurendra Kumar, Alex W. Schrader, Bhavay Aggarwal, Ali Ebrahimpour Boroojeny, Marisa Asadian,JuYeon Lee, You Jin Song, Sihai Dave Zhao, Hee-Sun Han & Saurabh Sinha

Nature Communications, September 2024

Paper Code

CoVA: Context-aware Visual Attention for Webpage Information Extraction

Anurendra Kumar‡, Keval Morabia‡, Jingjin Wang, Kevin Chen-Chuan Chang, Alexander Schwing

ACL 2022

Paper Code

Monoaural Audio Source Separation using Variational Autoencoders

Anurendra Kumar‡, Laxmi Pandey‡, Vinay Namboodiri

Interspeech 2018

Paper Code

Dirichlet latent variable model: A dynamic model based on dirichlet prior for audio processing

A Kumar, T Guha, PK Ghosh

IEEE/ACM Transactions on Audio, Speech, and Language Processing 2018

Paper Code

A Dynamic Latent Variable Model for Source Separation

A Kumar, T Guha, PK Ghosh

ICASSP 2018

Paper Code

CellWHISPER disentangles direct cell-cell communication from structural proximity

Anurendra Kumar, Felix Rivera Moctezuma, Bhavay Aggarwal, Nicholas Zhang, Ahmet F Coskun & Saurabh Sinha

Biorxiv 2026

Code

Intracellular spatial transcriptomic analysis toolkit (InSTAnT)

Anurendra Kumar, Alex W. Schrader, Bhavay Aggarwal, Ali Ebrahimpour Boroojeny, Marisa Asadian,JuYeon Lee, You Jin Song, Sihai Dave Zhao, Hee-Sun Han & Saurabh Sinha

Nature Communications, September 2024

Paper Code

CoVA: Context-aware Visual Attention for Webpage Information Extraction

Anurendra Kumar‡, Keval Morabia‡, Jingjin Wang, Kevin Chen-Chuan Chang, Alexander Schwing

ACL 2022

Paper Code

Monoaural Audio Source Separation using Variational Autoencoders

Anurendra Kumar‡, Laxmi Pandey‡, Vinay Namboodiri

Interspeech 2018

Paper Code

Dirichlet latent variable model: A dynamic model based on dirichlet prior for audio processing

A Kumar, T Guha, PK Ghosh

IEEE/ACM Transactions on Audio, Speech, and Language Processing 2018

Paper Code

A Dynamic Latent Variable Model for Source Separation

A Kumar, T Guha, PK Ghosh

ICASSP 2018

Paper Code

Vitæ

Full Resume in PDF.

Anurendra Kumar
anurendk@stanford.edu