Projects

A novel computational pipeline for species tree inference that combines Locus-of-Sequence (LoS) and Whole Genome (WG) data in a principled manner. LoS datasets offer high taxon sampling across tens of thousands of species but limited locus quality; WG data provides numerous high-quality loci but narrower taxon coverage. MAGNET bridges both worlds.

• Accurate & Interpretable — topology + branch lengths + phylogenetic uncertainty estimates
• Scalable — designed for tens of thousands of species using HPC parallelization
• Versatile — handles varying WG assembly quality (scaffold vs. chromosome-scale) and diverse LoS types (exons, UCEs)
• Expandable — incrementally incorporates new sequencing data without full de novo re-inference
• Broadly usable — largely automated with a user-friendly interface for non-experts

Reference free, Orthology free, Alignment free, Discordance aware Estimation of Species tree — a fully automated, end-to-end HPC pipeline for inferring species trees directly from raw genome assemblies.

• Achieved 176× speedup in tree generation for large-scale genomics datasets
• Published in PNAS 2025, featured on the journal cover; presented at ISMB 2024
• 2K+ conda downloads, 43 GitHub stars; actively used by researchers worldwide
• Collaborating with the Vertebrate Genomes Project (VGP) consortium and Scripps Institution of Oceanography

A GPU-accelerated and incremental extension of ROADIES for large-scale species tree placement. Rather than re-running full de novo inference when new genomes become available, ROADIES-Placer places new taxa onto an existing species tree incrementally — significantly reducing compute time and enabling continuously updated phylogenies as sequencing data grows. GPU acceleration further speeds up the core alignment and tree estimation steps for massive datasets.

An HLS-based open-source framework for creating FPGA accelerators for dynamic programming algorithms in bioinformatics — covering sequence alignment, homology search, basecalling, and more.

• Achieved 32× speedup and 20× faster development time than hand-coded HDL
• 15 diverse DP kernels implemented; compatible with AWS EC2 F1 FPGA instances
• Published at IEEE HPCA 2026, Sydney, Australia

• Parallelized Suffix Array construction on GPUs — 86–571× speedup over optimized CPU baselines for genomic read mapping
• Accelerated dense matrix multiplication using CUDA on NVIDIA K80/T4 GPUs; implemented Intel AVX2 vectorization
• Optimized large-scale solvers (Aliev-Panfilov) on the Expanse Supercomputer using MPI and C++

Improved the HLS4ML library for efficient ML hardware inference via High-Level Synthesis. Partnered with CERN's HLS4ML team to implement feature enhancements and optimizations on the DL-to-FPGA flow.