15 December 2015. Pharmaceutical chemists at University of California in San Francisco developed techniques for adapting computer vision, like those used in robotics, to early-stage drug discovery. The team led by UC-San Francisco’s Steven Altschuler and Lani Wu describe its discovery in yesterday’s issue of the journal Nature Biotechnology (paid subscription required).
Altschuler and Wu, with colleagues from University of Texas Southwestern Medical Center where they began their research, are seeking better tools for identifying new drugs that induce desired responses from cells, but work in different ways than known compounds. One of those improved tools is to rapidly organize and annotate molecules in compound libraries based on their biological functions and pathways. The authors envision creating drug discovery tools similar to those in genomics, where functions of genes are cataloged and made available for high-speed computerized screening.
Computerized screening is used today for drug discovery in facilities known as cores, but as Wu explains in a UC-San Francisco statement, their processes can be improved. “The problem is that screening results generally cannot be reused,” says Wu. “When you have a new biological target you want to hit with a drug, you have to go and screen the whole compound library again.”
To meet this objective, the team designed a new type of reporter cell, the kind used in drug discovery that changes appearance or behavior when reacting to a drug candidate. In this case the reporter cells, which the researchers call optimized reporter cell lines for annotating compound libraries, or Oracls, would need to distinguish among multiple common drug classes.
The team started with a 93 cell lines already associated with lung cancer and attached fluorescent tags to a random selection of genes from those cells. The researchers then treated the cell lines with 30 compounds from 6 classes of common cancer drugs, which as expected, caused different reactions among the lung cancer cells, changing their shape and fluorescence patterns. With computer-vision algorithms, the team was able to identify distinctive patterns in these reactions that indicated the type of drug used.
One reporter cell line in particular distinguished between the 6 different drug classes with 94 percent accuracy. As important, the analysis revealed common types of visible reactions in cancer cells to different drug classes. For example, one class of drugs known as histone deacetylase or HDAC inhibitors caused cells to form spikes, while another type of drug called mammalian target of rapamycin or MTOR inhibitors created a dim fluorescence pattern in the cells.
The researchers tested their Oracl computer-vision cell analysis techniques by screening more than 10,000 small-molecules acquired from compound libraries at other institutions, whose functions were unknown. In a single pass, their screens returned 106 molecules with effects similar to those exhibited in the initial round to identify the Oracl cells. Further tests showed 90 of those 106 molecules affected the same biological pathway. In addition, the screens revealed compounds from a number of classes of drugs not in the first analytical round, including some whose biological functions are still not known, that still produced characteristic reactions to the Oracl cells.
Altschuler and Wu filed for a patent on their techniques, and are scaling up their analysis and annotation process to cover hundreds of thousands of molecules. They believe Oracl can be used in early drug discovery stages to quickly and efficiently identify new candidates from drug classes not usually associated with the target biological pathways or patient populations.
“Currently the process takes billions of dollars over many, many years,” says Altschuler. “Wouldn’t it be nice to make that easier?”
Read more:
- NIH Funding New Robotics for Disabled, Children
- Crop Disease Images Collected, Available Via Smartphones
- Computer Model Provides Early Sepsis Alert
- Computer Model Predicts Protein Binding to DNA, RNA
- Computer Model Predicts Bacteria Mutations, Aids Drug Design
* * *
You must be logged in to post a comment.