The promise of AI and accelerated scientific discovery

By | 3 minute read | July 14, 2020

female scientist

Rapid drug discovery processes are critical to dealing with new viral outbreaks and epidemics, such as COVID-19. The traditional drug discovery pipeline is costly and time intensive. For a new drug to reach the market, it can take up to 10 years and cost as much as $2.6 billion. And, a vast majority of drug candidates fail at the testing stage … meaning valuable time and money is lost along the way.

Generative AI models have shown promise for speeding and facilitating the drug discovery process by automating molecule discovery. IBM Research’s Science for Social Good has been exploring this application for the last two years and has recently applied it to COVID-19.

The limitations and promise of using AI for drug discovery

Many challenges exist with using generative AI frameworks to accelerate the design of novel drug candidates. Typically, these generative frameworks are not efficient in handling design tasks with multiple discovery constraints and learning from limited labels, have limited exploratory capabilities, and require expensive model retraining to learn beyond limited training data.

The robust generative AI frameworks designed by our team, however, can overcome these challenges, and can assist in the creation of novel peptides, proteins, drug candidates, and materials. We combine and leverage deep learning, curriculum learning, generative modeling, and novel sampling and optimization methods to enable controllable generation of novel artifacts with desired properties.

Using this generative AI framework, for example, we’ve discovered two novel antimicrobial peptides with high broad-spectrum potency and selectivity with a higher success rate (10%) and faster development rate (48 days), compared to state-of-the-art methods (<1% and 2-4 years).

Road-testing AI on the current pandemic

We have applied this AI technology against three COVID-19 targets to identify more than 3,000 new small molecules as potential COVID-19 therapeutic candidates. IBM is releasing these molecules under an open license, and researchers can study them via a new interactive molecular explorer tool to understand their characteristics and relationship to COVID-19 and identify candidates that might have desirable properties to be further pursued in drug development.

The molecular explorer tool enables you to select a biological target and filter generated molecules by important characteristics, view related molecules and nearest match in PubChem, and see relationships among molecules.

And, given recent studies in zoonotic spillover are showing an increased likelihood of experiencing new outbreaks of diseases that transfer from animals to humans, this work is critical to rapid response if these occur.

A vision of accelerated discovery

Our vision for the future of accelerated discovery brings together AI researchers and pharmaceutical scientists to rapidly create next-generation therapeutics, aided by novel AI-powered tools.

As part of this vision, we’ve led the launch of the U.S. COVID-19 High Performance Computing Consortium, which is harnessing massive computing power in the effort to help confront the coronavirus. Further, to facilitate the mining of the vast troves of medical research and data that could provide relevant to COVID-19, we are offering a cloud-based AI research resource that has been trained on a corpus of thousands of scientific papers contained in the COVID-19 Open Research Dataset (CORD-19), prepared by the White House and a coalition of research groups, and licensed databases. This tool uses our advanced AI and allows researchers to pose specific queries to the collections of papers and to extract critical COVID-19 knowledge quickly.

Beyond COVID-19: Expanding the application of AI to other social needs

Beyond IBM Research’s response to COVID-19, Science for Social Good is exploring how to apply these same technologies across immediate, social needs. Other projects already completed include: using machine learning to hunt the Zika Virus; understanding humanitarian crises in real time; recognizing hate speech; applying AI to combat the opioid crisis; outlining optimal paths out of poverty; and identifying what works in global development. And this work is just the beginning.