7th DeepChem User Group Meetup

Bay Area DeepChem User Group
Bay Area DeepChem User Group
Public group

Verily Headquarters

269 East Grand Avenue · South San Francisco, ca

How to find us

Please park in the front parking lot. If anyone gets lost, please hone 415-515-1604 for directions.

Location image of event venue


The 7th DeepChem meetup will be hosted by Freenome at Verily's headquarters. Come by to learn the latest about DeepChem and about uses of machine learning in bio!


Imran Haque, CSO @ Freenome
Speech Abstract:
Learning problems in molecular biology and chemistry often have a similar total amount of data as large problems in other realms of machine learning -- TBs and up -- but a very different shape, with a small number of very wide instances rather than many narrow instances, and with rampant statistical confounding from intrinsic variability in underlying physical measurements. In this talk I will present case studies from our work at Freenome illustrating pitfalls of blind applications of machine learning to real-world genomics datasets, demonstrate the value of domain-specific analyses that have not yet been implemented in an end-to-end-learned fashion, and propose useful directions of research in fundamental DL methods to overcome these challenges.

Bharath Ramsundar:
Bharath did his PhD in computer science at Stanford University pn deep-learning for drug-discovery. Today, Bharath is focused on designing the decentralized protocols that will unlock data and AI to create the next stage of the internet.
Speech Abstract:
Bharath will be speaking about the release of DeepChem 2.1 and the new features it brings, such as uncertainty estimation, tensorflow eager integration, and early support for microscopy datasets.

Jo Varshney, DVM, Ph.D.
Co-founder and CEO, VeriSIM Life
She is a DVM and received her Masters. in Translational Pathobiology & Bioinformatics from Penn State University, focusing on gut microbiome analysis for colitis patients. She later pursued her Ph.D. in Comparative Oncology/ Genomics & CS from University of Minnesota-Twin Cities, where she identified targeted small compound for Osteosarcoma that is currently in Phase 1 trial. Jo also served as a Visiting Research Scientist at Genentech working on understanding the mechanisms of toxicities induced by antibody drug conjugates.

VeriSIM Life is building AI enabled biosimulation models.
VeriSIM will tackle one of the biggest obstacles of drug development: animal testing for drug development. Animal testing is slow, ethically questionable, and doesn't act as much of a filter: 92% of all drug candidates that pass this preclinical testing never make it to market. VeriSIM's solution is to create disease-specific biosimulation models, which allow researchers at pharma companies to model in software how a drug will interact in animals and bring more personalized human clinical trials.

Sam Schoenholz:
Sam is a research scientist at Google Brain. Sam's research interests lie at the intersection of science and machine learning. He has used deep learning to predict the quantum mechanical properties of small molecules, to understand dynamical arrest in glasses, and most recently to improve the state of peptide-spectra matching. Simultaneously, he uses methods from statistical physics to better understand deep learning models. Sam has a PhD in Physics from the University of Pennsylvania and a BA in Physics & Mathematics from Swarthmore College.

Title: Peptide-Spectra Matching from Weak Supervision
Abstract: As in many other scientific domains, we face a fundamental problem when using machine learning to identify proteins from mass spectrometry data: large ground truth datasets mapping inputs to correct outputs are extremely difficult to obtain. Instead, we have access to imperfect hand-coded models crafted by domain experts. In this talk, we apply deep neural networks to an important step of the protein identification problem, the pairing of mass spectra with short sequences of amino acids called peptides. We train our model to differentiate between top scoring results from a state-of-the art classical system and hard-negative second and third place results.