Amir Shanehsazzadeh
I am an AI Scientist at Absci. I received my Bachelor's in Mathematics and Master's in Statistics from Harvard University. I am a graduate of Upper Merion High School in King of Prussia, PA. This website encompasses my previous and ongoing projects and experiences.
Experience
AI Scientist
I work on the AI team at Absci where I use state-of-the-art machine learning methods to model and design antibodies with the eventual goal of de novo design.
Machine Learning Associate Scientist
I worked on the Machine Guided Design team at Dyno where I designed ML models and built infrastructure for in-silico AAV library design processes.
Research Intern
I worked with David Belanger and David Dohan as part of the sequin (computational biology, protein and DNA sequence design team). I applied Transformer (BERT) models to biological sequence design. I developed a model for protein family prediction from fixed-length vector representations of proteins using contextual lenses from NLP. I first-authored two accepted workshop papers and open sourced the project.
Computational Sciences Research Intern
I worked with Paul Maragakis and Hunter Nisonoff as a summer intern to build DESSEQUENCE, a tool that measured the performance of molecular dynamics simulations. DESSEQUENCE relies on an underlying deep neural translation model between protein sequences and binary labels, which represent the resolution of the amino acid residues in the sequence. If interested, please contact me for more information.
Computational Biology and Data Science Researcher
I worked with Dr. Roland Dunbrack at Fox Chase Cancer Center to study and develop new approaches to clustering data, particularly protein structural conformations. I developed DIHNOSIR: Density-Independent High Noise Optimized Sortingy by Iterative Reduction. DIHNOSIR combines numerous iterations of the famous clustering algorithm DBSCAN with a graph-theoretic approach to remove merged clusters produced by DBSCAN and piece together a more accurate result. The clustering of these structural conformations allows for improved protein structure, design, and validation. You can find a paper I wrote on DIHNOSIR here.