Insilico offers AI-based chemistry to coronavirus response
Insilico Medicine Inc. became the second AI company this week to contribute the speed of its platform to biopharma’s mass mobilization against the 2019-nCoV crisis.
As the debate continues about whether AI can transform medicinal chemistry versus just speed it up, outbreaks like the 2019-nCoV could be a proving ground for companies that believe their computational approaches can find compounds in a fraction of the time of traditional methods.
Insilico announced Thursday it used its computational chemistry platform, dubbed GENTRL (generative tensorial reinforcement learning), to generate small molecule structures predicted to bind the 2019-nCoV 3C-like protease.
The company said it first selected the target on Jan. 28, and had generated and published the structures by Feb. 5. Insilico is collaborating with an undisclosed synthetic chemistry company to synthesize and test up to 100 of the compounds, and has invited others to synthesize, test and provide feedback on the remaining structures, which the company has made available on its website.
GENTRL incorporates machine-learning algorithms known as generative adversarial networks (GANs), which use input data as a template to guide the design of de novo outputs (see “Insilico, WuXi Unveil One Piece of AI Pipeline”).
In a preprint submitted to the bioRxiv server, Insilico described the three approaches it used to generate candidate structures: one based on the crystal structure of a coronavirus 3C-like protease, another based on the structure of a co-crystallized ligand for the protease, and a third based on computational modeling of a homologous SARS protein.
On Tuesday, BenevolentAI published a Lancet correspondence saying its target and compound discovery technology nominated repurposing candidates with the potential to limit viral entry (see “2019-nCoV Drug Discovery: BenevolentAI Deploys Machine Learning”).
Further analysis of the coronavirus crisis can be found at https://www.biocentury.com/coronavirus. The 2019n-CoV content is free to all who visit the site.