Data imputation for drug repositioning

December 16, 2022
12:00pm ET
Cummings 076
Speaker: Di Zhou
Host: Donna Slonim


Quals talk:

The huge investment yet low profitability of the traditional drug development process makes the risk of novel drug discovery ascend drastically. Drug repositioning, which refers to the process of investigating existing drugs to apply to new therapeutic purposes, provides a new approach for drug development. The Connectivity Map (CMap) project and the Library Integrated Network-based Cellular Signatures (LINCS) project are the two main data repositories of transcriptomic profiles of an abundant number of drug exposures to different cell lines, and they improve drug repositioning exponentially. However, experimentally gathering expression profiles of all combinations of different drugs and cells is a costly process, which leads to around 70% of missing cell-drug combinations in these datasets. To better do drug repositioning, we therefore need to impute the missing expression profiles. Previous studies showed that using a neighborhood-based collaborative filtering method to impute missing profiles can improve drug repositioning. However, this collaborative filtering method considers the similarity of drugs across all cell types, which may contain noise because drug-induced expression profiles may have cell-type specific patterns. In this talk, we will introduce the biclustering method, which considers the effects of both drugs and cell types, to better cluster expression profiles of certain drugs and cell line combinations, hence helping improve data imputation and improving drug repositioning.

Research area: Computational biology

Please join meeting in Cummings 076.

Zoom is not available for this event; please disregard the Zoom information included in the email.