Protex, natural language for protein discovery

Protein-based drugs are a massive, decabillion market. They're undertapped and the number of possible proteins is more than the number of atoms in the universe. That said, sifting through so many proteins is incredibly challenging. Protex lets you take an English phrase (something like, "I want a protein that uses the ubiqitin pathway"), and search through large protein databases for therapeutic candidates. Protex won second place and the ChromaDB prize at the July 2023 Scale AI hackathon. Demo coming soon!

   Jane, a Q&A for proteomics annotation

Spatial proteomics experiments tag proteins in cells with markers, then images the markers to determine the location of the proteins. This is a powerful technique, but because everyone uses different markers and imaging techniques, it's difficult to impute expert knowledge when hand-labeling cells. Jane is a question-answering system given access to Janeway's Immunology and the expert curated marker information from the Human Reference Atlas. It can be used to efficiently train human annotators and provides an easy way to access esoteric knowledge.

   Deep learning cell segmentation/identification with spatial proteomics data

To be added on publication.

   Data compression for quantum machine learning

One really exciting quantum computing application is to combine machine learning and quantum algorithms (quantum machine learning). Modern quantum computers are in their infancy and hardware constrained, so so it's important to find useful algorithms that can be successfully run on relatively crude quantum hardware. This project introduces several such algorithms, benchmarks them on image classification tasks, and introduces a quantum dataset with images compressed in a format useful for future validation of quantum algorithms.

   Time evolution on near-term quantum hardware

Most useful scientific phenomena are dynamic, meaning they change over time. This is true of chemical reactions and atom-photon interactions, both of which have massively funded applications ranging from drug discovery to telecommunications. Quantum computers can theoretically simulate the true behavior of these processes, but to tackle these problems using quantum devices, we need algorithms that can run on unreliable hardware and be robust against noise (that's the current quantum hardware status). This project introduces two algorithms designed for time evolution on standard systems in condensed matter physics, and benchmarks both algorithms on supercomputing clusters and on one of IBM's quantum processors.