Trustworthy Deep Learning for Biology, Medicine, and Agriculture

My lab works across drug discovery, protein structural analysis, computational biology, computational pathology, and precision agriculture. Our projects range from analyzing pathology images and integrating them with genomic data to studying protein structures and small molecules for drug development, as well as applying computational methods to agricultural challenges.

Cal Poly Pomona students who want to do research with the lab can write to skosaraju@cpp.edu.

Where cholesterol binds

PIEZO2 channel with three cholesterol molecules at predicted binding sites — Three cholesterol-binding sites CholBindNet predicted on the PIEZO2 channel, shown in magenta. The sites had high, intermediate and low occupancy in molecular dynamics simulations.

Two helix structures with cholesterol and the residues the model weighted most — The residues the model weighted most heavily for two known structures, with cholesterol in teal. Pointing at the atoms behind a prediction is the part that makes it checkable.

From CholBindNet (Communications Chemistry, 2026), with the Luo Lab.

What the models see

StomachHipoMap 2022 · survived 1 month

HipoMap patch probability map over a stomach adenocarcinoma slide, patient who survived 1 month

StomachHipoMap 2022 · survived 23 months

HipoMap patch probability map, patient who survived 23.3 months

StomachHipoMap 2022 · survived 53 months

HipoMap patch probability map, patient who survived 53.1 months

LungDeepPATHO · ALK 2025

DeepPATHO ALK-probability heatmap over a lung resection slide

LungDeepPATHO · ALK 2025

DeepPATHO ALK-probability heatmap over a second lung resection slide

GastricDeep-Hipo 2020

Deep-Hipo probability map over a multi-fragment gastric slide

Patch-level probability maps laid over the tissue, from HipoMap (Scientific Reports, 2022), ALK screening (npj Digital Medicine, 2025) and Deep-Hipo (Methods, 2020). Red is a high probability, blue a low one; outlines mark annotated regions.

Research

All research

Evidential deep learning for medical image analysis

Models for whole-slide pathology images that give a prediction along with how much evidence supports it. Current work screens H&E-stained lung cancer slides for ALK rearrangements.

Tumor patches with attention maps from the ALK screening model

Multi-omics data integration for prognosis

Survival and outcome prediction from gene expression, DNA methylation, copy number and pathology images, built around biological pathways so a result can be read as a pathway effect.

Pathway-pathway attention heatmap from PINT

Interpretable protein sequence and structure analysis

Predicting enzyme function, cholesterol-binding sites and protein-compound interactions in a way that points to the residues or atoms involved. With the Luo Lab, also molecular dynamics and molecule generation.

PIEZO2 channel with cholesterol-binding sites predicted by CholBindNet

Demographic-specific electronic health records analysis

How outcomes and model behavior differ by race and sex, from dementia care costs in Medicare and Medicaid records to sex-specific signals in transcriptomic data.

Error distributions by race for telehealth advance care planning cost models

Principal investigator

Sai Chandra Kosaraju, Ph.D.

Assistant Professor of Computer Science

Sai Chandra Kosaraju joined the Department of Computer Science at Cal Poly Pomona in Fall 2024. Dr. Kosaraju's research is on interpretable and evidential deep learning for health and biology: diagnosing disease from medical images, prognosis from human genomics, race- and sex-specific analysis of health records, and protein-ligand binding.

CS 4210, Machine Learning and Its Applications (Fall 2024)
CS 4990, Generative Deep Learning and Applications (Spring 2025)

Recent work

All publications

2026Docking-score landscapes shape active-learning performance across Vina, Glide, and SILCS
J. Comput. Aided Mol. Des.

2026PINT: Pathway-pathway interactions for predicting interpretable clinical outcomes from gene expression
bioRxiv

2026PIMO: pathway-based interpretable multiomics interactions for multiomics integration
Bioinformatics

2026Fairness-aware supervised hierarchical contrastive semantic learning for sexual dimorphism analysis
Bioinformatics

2026Slide-Omics: An Interpretable Bi-Directional Attention Framework for Integrating Multi-Omics and Pathology in Cancer Survival Analysis
ACM-BCB

Support

The lab's work is supported by:

ARI Campus Grant, 2026–2027

Agricultural Research Institute, the CSU's applied agricultural research consortium.

cpp.edu

ARI-NextGen Fellowship

Student fellowships from the CSU Agricultural Research Institute, funded by the USDA NIFA NextGen program.

calstate.edu

WesternU Molecular Computing Core

WMCC provides the high-performance computing behind the lab's protein and small-molecule work.

research.westernu.edu

TIDE

Technology Infrastructure for Data Exploration: NSF-funded GPU computing for CSU researchers, hosted at San Diego State.

tide.sdsu.edu

Cal Poly Pomona Foundation

Thanks as well to the labs and pathologists the lab works with, listed on the People page.