Surag Nair

The paper introduces Decima, a model that predicts cell type- and condition-specific gene expression from DNA sequence using single-cell RNA sequencing data.

This paper introduces CompBioBench, a benchmark for evaluating agentic systems in computational biology, showing strong performance across diverse tasks.

This study designs cell type-specific AAV promoters using deep learning, outperforming rational approaches in mouse retinal cells and showing promise for human applications.

Nona is a multimodal masking framework that unifies sequence-to-function models, language models, and generative models for functional genomics, improving predictions and revealing privacy vulnerabilities in ATAC-seq data.

The paper creates a single-cell multiome atlas of RPE and choroid tissues to identify pathogenic variants and mechanisms in AMD using HiChIP, Activity-by-Contact, and STARR-seq.

Learning and removing Tn5/DNase-I sequence bias lets a small base-resolution CNN (ChromBPNet) recover TF motifs and footprints from accessibility data and rival Enformer at predicting chromatin QTL effects.

gReLU provides a framework for DNA sequence modeling and design, including tools for training, evaluating, and applying pre-trained models to predict and create regulatory elements.

The paper uses deep learning to analyze chromatin accessibility and gene expression during reprogramming, revealing how TF concentrations and motif syntax influence cis-regulatory networks.

The paper investigates chromatin accessibility changes in neurogenic niche cells during aging, identifying defects in NSC adhesion and migration and proposing ROCK inhibition as a potential therapeutic strategy.

This study uses single-cell chromatin accessibility analysis to identify developmental trajectories and non-coding mutations associated with congenital heart disease.

AIMHIGH identifies novel disease-causing genes by analyzing inbred mouse strain data and machine learning-driven literature-based discovery.

The paper uses a single-cell multiome atlas of the human retina and deep learning to identify causal noncoding variants in eye diseases.

This paper introduces Yuzu, a method that accelerates in-silico saturation mutagenesis by using compressed sensing, reducing computation time for convolution-based models.

The paper introduces a genome browser track for visualizing dynamic DNA sequence features at single nucleotide resolution using machine learning-derived scores.

fastISM accelerates in-silico saturation mutagenesis for convolutional neural networks by 10x, making it feasible for large-scale sequence analysis.

The paper introduces MPRA-DragoNN, a CNN model to predict regulatory DNA activity from MPRAs, identifies TF motifs, and explores variant effects in lipid-associated loci.

The paper predicts genome-wide chromatin accessibility across cell types by integrating DNA sequence and gene expression data using neural networks.