Their findings, which appeared in preprint November 8 in the journal bioRxiv, are slated for presentation at the 36th Conference on Neural Information Processing Systems, held this week in New Orleans.

Drug discovery typically involves identifying a biological target associated with a disease, then screening libraries of thousands of chemical compounds that might help treat that disease. The most promising candidates advance to laboratory and clinical testing.

The UT Dallas-Novartis team used topological data analysis to screen thousands of possible drug candidates virtually and then narrowed the compound candidates to those best fit for further testing. While virtually screening libraries for chemical compounds is not new, the researchers contend that their approach outperforms others on large data sets.

The team framed the virtual screening process as a topology-based graph ranking mathematical problem, which characterizes each molecular compound by the shape of its underlying physical substructure -- its topology -- as well as its physical and chemical properties. A unique "topological fingerprint" for each compound helps rank it according to how well it fits the desired properties.

Next in the process is molecular property prediction, including scoring compounds for their solubility in water -- a property critical to a drug's efficacy in the body. The new method then ranks compounds based on how likely they are to work, helping companies avoid dead ends.

"The advantage of our algorithm is that it could screen about 100,000 compounds in a couple of days, which is much faster than other methods," co-author Baris Coskunuzer, PhD, professor of mathematical sciences at UT Dallas, said in a statement.