Vaibhav Rajan

PatternExtract: A facile, scalable pipeline for point pattern generation from spatial imaging data

PatternExtract generates spatial point patterns from pathology images and cell coordinates using a two-kernel segmentation method and automated classification, validated on lymphoma data.

Neural Collective Matrix Factorization for Integrated Analysis of Heterogeneous Biomedical Data

This paper introduces NCMF, a neural collective matrix factorization method for integrating heterogeneous biomedical data, showing improved performance in gene-disease and adverse drug event prediction.

scMoMaT: Mosaic integration of single cell multi-omics data using matrix trifactorization

scMoMaT integrates single-cell multi-omics data under mosaic integration using matrix tri-factorization, uncovering modality-specific biomarkers for improved cell type annotation.

DruID: Personalized Drug Recommendations by Integrating Multiple Biomedical Databases for Cancer

DruID integrates drug efficacy predictions with genetic and pharmacological data to provide personalized cancer drug recommendations, outperforming existing methods on cell line data.

ASTER: A Method to Predict Clinically Actionable Synthetic Lethal Interactions

ASTER identifies clinically actionable synthetic lethal gene pairs using statistical hypothesis testing on gene expression data.

Maximum Likelihood Reconstruction of Ancestral Networks by Integer Linear Programming

The paper presents an Integer Linear Programming approach for optimal ancestral PPI network reconstruction under the DMC model, showing improved likelihood and agreement with biological evidence.

Mining Pathway Associations from Networks of Mutual Exclusivity Interactions

The paper introduces a new network mining algorithm to identify pathway-centric interactions in cancer data, revealing dense between-pathway interactions with therapeutic potential.

Clustering using Automatic Differentiation: Misspecified models and High-Dimensional Data

The paper investigates clustering with Gaussian Mixture Models under misspecification and high-dimensional data, showing EM outperforms GD in misspecification but GD is better for high-dimensional data, and introduces a penalty term using AD for improved clustering.

Predicting Synthetic Lethal Interactions usingHeterogeneous Data Sources

The paper develops a method to predict synthetic lethal interactions by integrating heterogeneous data sources, outperforming existing methods with minimal feature engineering.

Deep Augmented Multiview Clustering

This paper introduces a deep learning method for multiview clustering that co-clusters matrices and identifies cross-matrix cluster associations, with initial results on a clinical dataset.