The fight against COVID-19 has seen a call to action to all facets of scientific discovery. This includes AI which stands to transform how we react to the pandemic and empower our scientists and public policymakers with newer and more powerful tools. As part of our corporate responsibility and AI for Good initiatives, IBM has been at the forefront of this movement. This 1-hr talk will highlight members of our community who have risen to this challenge, in the context of AI research and the NeurIPS audience. This includes data analysis for non-pharmaceutical interventions (NPIs) world-wide to help in policy decisions, research in natural language processing, and molecule discovery to advance scientific research. The WNTRAC Open Challenge deals with AI-based data generation and analysis of non-pharmaceutical interventions to track COVID-19.
The Worldwide Non-Pharmaceutical Interventions Tracker for COVID-19 (WNTRAC) is a publicly available comprehensive dataset consisting of more than 6,000 NPIs implemented worldwide since the start of the pandemic. IBM Research has created a system that leverages DL technologies applied to Wikipedia pages for generating the data. The team has illustrated various ways to leverage the data for predicting disease spread and offers mechanisms to explore the causal effect of different interventions on the pandemic. Researchers are invited to explore the data.
Natural Language Processing for COVID Literature Search and Q&A This part of the talk will focus on the use of natural language processing to help scientists accelerate their discoveries as well as answer your COVID-19 questions from the scientific literature. It will demonstrate how to derive insights from a large corpus of papers and open datasets through Deep Search and Q&A.
Molecule Explorer using Generative AI This part of the talk will show how generative AI frameworks have been used to help researchers generate potential new drug candidates for COVID-19, applied to three COVID-19 targets to produce>3500 novel molecules and their attributes in the molecule explorer platform under an open license. We will talk about how generative AI techniques required to be “controllable” and be able to learn from limited labels and generalize to unseen contexts such as a novel viral protein. Such AI methods lead a path toward faster generation and comprehensive virtual screening of new and optimal candidate molecules and show promise for accelerating molecule and material discovery, which is critical for responding to unprecedented crises like the COVID-19 pandemic.