Research

Research Experience

Johns Hopkins University, Dr. Jeffrey Gray’s Lab

Protein Modeling Research Intern — June 2025 – August 2025

• Fine-tuned DFMDock (AI model for protein-protein docking) with ranking loss function and data augmentation
• Showed that fine-tuned model outperforms state-of-the-art methods for ranking protein-protein complex predictions
• Fellowship: NSF Rosetta Commons REU intern (3.7% acceptance rate)

Publications & Presentations:

• Zhu, R., Xu, D., Chu, L., & Gray, J. (2025). Improving scoring functions for protein–protein docking with LambdaLoss. Machine Learning in Structural Biology (MLSB) Workshop [Paper]
• Presented at Molecular Machine Learning (MoML) Conference and Summer RosettaCon

Harvard Medical School, Dr. Marinka Zitnik’s Lab

Machine Learning Research Intern — June 2024 – Present

• Developed tools and case study for ToolUniverse, an ecosystem for building AI scientists with 600+ tools
• Helped train ToolRAG model for large-scale tool retrieval from ToolUniverse
• Developed training and evaluation data for TxAgent, an AI agent using ToolUniverse and multi-step reasoning to reason about drugs and personalized medicine; built Gradio platforms for scientists evaluating TxAgent
• Developed LLM-powered evaluation system for CURE-Bench, an international NeurIPS competition co-organizing to benchmark AI models on therapeutic reasoning
• Fellowships:
  • Summer Institute for Biomedical Informatics (SIBMI) fellow (10% acceptance rate)
  • Kempner Undergraduate Research Experience (KURE) fellow

Publications & Presentations:

• Gao, S., Zhu, R.*, Sui, P.*, Kong, Z.*, Aldogram, S.*, et al. (2025). Democratizing AI scientists using ToolUniverse. arXiv preprint arXiv:2509.23426 [Paper, Website, GitHub]
• Gao, S., Zhu, R., Kong, Z., et al. (2025). TxAgent: An AI agent for therapeutic reasoning across a universe of tools. arXiv preprint arXiv:2503.10970 [Paper, GitHub]
• Gao, S., Zhu, R., et al. (2025). CURE-Bench: Competition on reasoning models for drug decision-making in precision therapeutics. NeurIPS Competition Track
• 2nd Place, Best Poster — Harvard Dept. of Biomedical Informatics Science Day (2025)

Massachusetts General Hospital, under Dr. Abhiram Bhashyam

Computational Clinical Research Intern — June 2023 – Present

• Analyzed time-series hand-drawing data of 130+ patients to identify pressure features linked to impaired hand function

Publications & Presentations:

• Laane, C.*, Zhu, R.*, Chandra, J., et al. (2024). Hand and wrist immobilization affects pressure in complex spiral-based drawings. Tufts Biology Research Symposium

Brigham and Women’s Hospital, Dr. Tracy Young-Pearse’s Lab

Molecular Biology Intern — February 2023 – May 2024

• Applied differential expression and gene set enrichment analyses to identify SMOC1 target in Alzheimer’s disease; used wet lab assays (e.g., Western Blot) to probe its mechanism; also studied autophagy proteins in Alzheimer’s
• Fellowship: Harvard Program for Research in Science & Engineering (PRISE) fellow

UCLA, Dr. X. William Yang’s Lab, under Dr. Peter Langfelder

Bioinformatics Intern — June 2021 – August 2021

• Developed novel methods for estimating mammalian brain age using elastic net regression and gene expression data
• Identified genes upregulated/downregulated with age using differential expression and functional enrichment analyses
• Fellowship: Research Science Institute (RSI) Scholar (3.7% acceptance rate)

Cabrini University, Dr. Vinayak Mathur

Bioinformatics Intern — June 2020 – December 2021

• Used bioinformatics methods to identify 94 virus-derived prophages in Acinetobacter pittii bacteria, which contained 37 resistance genes and 47 virulence factors that could influence pathogenicity

Publications & Presentations:

• Zhu, R., & Mathur, V. (2022). Prophages present in Acinetobacter pittii influence bacterial virulence, antibiotic resistance, and genomic rearrangements. PHAGE, 3(1), 38–49 [Paper]
• Mancini, K., Cao, I., Ge, C., Mu, E., Zhu, R., & Mathur, V. (2024). Investigating the role of phage mediated HGT in increasing bacterial virulence. Bios, 95(2), 73–78 [Paper]
• Presented at Harvard Microbial Science Initiative 20th Anniversary Symposium (2023) and NEMPET Conference (2021)

Course Projects

Harvard Geometric ML (AM220)

January – May 2025

Designed a method to construct phylogenetic trees in hyperbolic space using protein language model embeddings.

MIT Advanced NLP (6.8610)

October – December 2024

Used contrastive methods and protein language models for functional edits in protein engineering. Paper accepted to the 2025 ICLR Learning Meaningful Representations of Life workshop (Tiny Paper Track).

Publication: Pan, R.*, Zhu, R.*, Lakshman, V., & Qu, F. (2025). Mutagenic: An embedding-based approach to protein masking for functional redesign. ICLR LMRL Workshop [Paper]

Evolved 2024 (Bio × ML Hackathon)

October 2024

Built a graph-based AI model to answer conjunctive logical queries on a biomedical knowledge graph. Constructed training data from PrimeKG and test data from TWOSIDES polypharmacy graph. Won Scale Medicine prize.

MIT Drug Discovery (20.201)

September – December 2024

Constructed a computational pipeline (target ID, docking screen, ADMET optimization) to identify a novel molecular glue for degrading DUX4, the cause of facioscapulohumeral muscular dystrophy (FSHD). Implemented ADMET-AI model for pharmacokinetic property optimization.