Work

Papers

Multivalent binding model quantifies antibody species from systems serology Armaan A. Abraham, Zhixin Cyrillus Tan, Priyanka Shrestha, Emily R. Bozich, Aaron S. Meyer (2024). Submitted to PLOS Computational Biology.

Integrative, high-resolution analysis of single cells across experimental conditions with PARAFAC2 Andrew Ramirez, Brian T. Orcutt-Jahns, Sean Pascoe, Armaan A. Abraham, Breanna Remigio, Nathaniel Thomas, Aaron S. Meyer (2024). Preprint.

Projects

• Unaligned Low-rank Tensor Regression with Attention (ULTRA)
  A tensor-learning algorithm I invented for automatically constructing mechanistic explanations of arbitrary sample metadata (e.g. patient disease status) from scRNA-seq.
• Protein design with RL from AI feedback (github)
  Using Alphafold2 to train a reinforcement learning agent to generate peptide sequences conditioned on fuzzy structural design criteria.
• Reinforcement learning of human-interpretable, multimodal tasks from AI feedback (github)
  Using a text+image LLM to train agents to perform arbitrary tasks that can be specified with human-interpretable text and images.
• Feature superposition, cooccurrence and output influence in LLMs: a sparse autoencoder-based analysis (github)
  Examining the effect of feature cooccurrence and downstream output influence on feature superposition.