Tech & Health

February 13, 2014

Informatics tools link protein structure, drug action: speaker

Russ Altman, M.D., Ph.D., professor of Bioengineering, Genetics and Medicine at Stanford University, posed an intriguing question at last week’s Discovery Lecture: is it possible to predict drug side effects from the 3D-structure of proteins?

Russ Altman, M.D., Ph.D., right, of Stanford University, with Vanderbilt’s Kevin Johnson, M.D., following Altman’s recent Flexner Discovery Lecture. (photo by Susan Urmy)

Russ Altman, M.D., Ph.D., professor of Bioengineering, Genetics and Medicine at Stanford University, posed an intriguing question at last week’s Discovery Lecture: is it possible to predict drug side effects from the 3D-structure of proteins?

The idea sounds “ridiculous,” Altman said, because “there’s a lot of biology that happens — a lot of biology — between having a drug bind a target and a clinical phenotype.”

But using informatics tools they developed, Altman and Tianyun Liu, Ph.D., in his group found an interesting correlation between the 3D-features of protein drug-binding “pockets” and predicted side effects.

“I am surprised that a 3D-structural based method can even come close to doing that,” Altman said.

Over the last 15 years, Altman and his colleagues have developed informatics tools to mine data in existing protein 3D-structure databases. Their FEATURE algorithms turn theoretical “balls” of protein structure into vectors of numbers, which allows protein microenvironments to be compared using computational methods.

Altman described four drug-related projects using the FEATURE system:

• PocketFEATURE looks for similarities in drug-binding pockets to discover new uses for existing drugs (drug repurposing)

• DrugFEATURE evaluates the “druggability” of a protein pocket

• FragFEATURE dissects protein pockets for their binding preferences for fragments of molecules

• SE-FEATURE uses protein pockets to predict side effect profiles

“The 3D-structure database is an amazing resource that we’ve been very excited to use for data mining approaches for pocket similarity, druggability, predicting fragment binding and then this kind of crazy, and potentially useful, side effect prediction,” Altman said.

“The FEATURE vector representation has gotten us much farther than we ever thought it would take us.”

Altman is excited about the possibility of using similar informatics strategies to look at protein pockets that have been mutated in cancer or in genetic diseases.

“Can we see how the pocket (has changed) and how that changes its specificity for drugs, so that we can do personalized drug repurposing,” he said.

Altman is the principal investigator of the Pharmacogenomics Knowledgebase (PharmGKB), a database of human genetic variation and how it impacts drug response. He also directs the Helix Group at Stanford, which creates and applies computational tools to solve problems in biology and medicine.

The Department of Biomedical Informatics sponsored Altman’s lecture. For a complete schedule of the Flexner Discovery Lecture series and archived video of previous lectures, go to www.mc.vanderbilt.edu/discoveryseries.