Yash Chainani Email & Phone Number

Emeryville, California, United States

• Developed one of the first fully integrated retrobiosynthesis software tools to autonomously suggest the biosynthesis and semi-synthesis of small-molecules using both multifunctional and monofunctional enzymes as well as synthetic chemistry.• Deployed previously trained XGBoost and message-passing GNN models to accelerate the design of feasible biosynthetic pathways.• Built and used a molecular docking pipeline with tools, such as DiffDock, GNINA, SMINA, posebusters, and py3Dmol to aid synthetic biologists in the computational protein design of potentially promiscuous enzymes.• Designed a microservice architecture to deploy all resulting software on the web for easy use by synthetic biologists. Docker was used to containerize a frontend GUI (streamlit), a backend (Django + retrobiosynthesis software), and a postgreSQL database. Orchestration of all containers was achieved using docker swarm and load balancing was achieved with Celery and Redis.

Evanston, Illinois, United States

• Synthetically generated over 11 million infeasible enzymatic reactions to circumvent the lack of negative data in the literature by strategically considering alternate reaction sites on substrates within the approximately 30,000 known reactions stored in publicly available metabolic databases.• Trained XGBoost binary classification models on resulting datasets to predict the feasibility of novel reactions. Models were trained using GPUs on high-performance computing clusters and in a distributed fashion via Ray. • Built an end-to-end data processing pipeline with RDKit to sanitize raw reaction data and create digital fingerprints of processed reactions that were ultimately assembled into feature vectors for machine learning.• Deployed this data pipeline to perform stratified train/ test/ validation splits while removing any duplicated feature vectors to minimize leakage between sets. Model hyperparameters were optimized on the validation set via a Bayesian hyperparameter optimization procedure.• Performed synthetic over-sampling of the minority class via the synthetic minority over-sampling (SMOTE) technique available in Scikit-learn so as to mitigate any class imbalance effects.• Benchmarked final XGBoost models trained on both fingerprint-based and description-based reaction feature vectors against existing models in the literature and demonstrated that our models perform better across all key metrics of AUPRC, F1, precision, and recall.• Upgraded ML models even further by training deep learning hybrid GNN-LSTM models in which small-molecules were featurized as graphs and after several message-passing layers, graph embeddings of molecules were sequentially passed into an LSTM model with a final sigmoidal layer for binary classification. PyTorch was used in training these deep learning models.

Berkeley, California, United States

• Improved existing experimental apparatus and procedures to accurately perform adsorption measurements of gaseous hydrocarbons (methane, ethane, 2,2-dimethylbutane) in various zeolite frameworks and molecular sieves at very low pressures close to vacuum• Developed a robust algorithm in Python to invert experimentally obtained adsorption data and elucidate likely underlying distributions of crystallite radii present in various zeolite samples by statistically discriminating between several potential radii distributions• Executed such algorithms by utilizing Lawrence Berkeley National Laboratory's high performance computing resources • Collaborated with research scientists from Chevron to perform adsorption measurements on proprietary zeolite catalysts developed at Chevron and communicated research findings back to Chevron scientists• Presented research findings at undergraduate research fairs and group meetings. Wrote an honors' on the techniques and algorithms researchers can employ to accurately conduct adsorption measurements in porous materials

Berkeley, California, United States

Berkeley, California, United States

Berkeley, California, United States

- Performed Pulse Layer Deposition (PLD) of various metals onto SrIO3 substrates and characterized resultant samples using Atomic Force Microscopy (AFM) and X-ray Diffraction (XRD)- Collaborated with a graduate student and a postdoctoral researcher to analyze changes in substrate surface after PLD- Communicated research findings to the laboratory’s Principal Investigator (PI) with recommendations for future experiments

Doctor Of Philosophy - Phd, Chemical And Biomolecular Engineering

Co-advised by Professors Keith Tyo and Linda Broadbelt to study bioinformatics Tools used frequently: RDkit, Tensorflow, Scikit learn.

Bachelor Of Science - Bs, Chemical And Biomolecular Engineering

Activities and Societies: Cal Rotaract, Robotics and Engineering for Youth, American Institute of Chemical Engineers UC Berkeley.