Cancer drug discovery does not suffer from a dearth of targets-instead it is in need of a way to prioritize target selection. Now, a team from The Institute of Cancer Research has created a computational algorithm that hones in on targets with strong biological validation that also are predicted to be druggable.1 The approach may provide a basis for systematically picking cancer targets.

Currently, decisions about cancer target selection are usually based on "gut feeling," said Stephen Friend, president, cofounder and director at Sage Bionetworks and former SVP and franchise head for oncology research at Merck & Co. Inc. "The calculus of how people weigh certain aspects of the decision-am I looking for strong valid biology, or does druggability trump all-is almost done as if we were alchemists."

A team led by Paul Workman and Bissan Al-Lazikani wanted to develop a platform to objectively compare the strengths and weaknesses associated with potential cancer targets.

Workman is deputy chief executive of The Institute of Cancer Research (ICR), head of ICR's division of cancer therapeutics and director of ICR's Cancer Research UK Cancer Therapeutics Unit. Al-Lazikani leads the computational biology and chemogenomics team in both the unit and the Division of Cancer Therapeutics at ICR.

"The massive and continual daily dump of cancer data far exceeds our ability to do detailed wet biology follow-up. The status quo is that the selection of targets to work on starts from an initial choice that is highly biased, such as which pathways or targets are familiar" to a given researcher, said Workman. "We felt there was a real need to be more unbiased and systematic and to do assessments rapidly and at scale at the most critical stage of drug discovery-choosing what targets to work on at the beginning."

"This is very important as each target can take a year or more to validate-or not-and the failure rate is high," Workman added.

As proof of principle, the computational approach developed by the ICR team focused on targets previously implicated in cancer and integrates multiple types of information to estimate the likelihood that a target will be druggable. The information includes target class, small molecule bioactivity data for the target that has been curated from published literature and analysis of potential small molecule binding sites based on experimentally determined protein structures or homology models.

From a list of 479 genes known to be altered in cancer, the method identified 29 oncogenes and 16 tumor suppressors that were predicted to be druggable but for which few or no small molecule ligands have yet been reported. This result suggests that these targets are relatively unexplored but potentially tractable for small molecule drug discovery.

The method was published in Nature Reviews Drug Discovery.

"The factors considered in this approach are ones that have been debated at the initiation of projects in pharma for many years. Unfortunately, these debates are often ad hoc in terms of the context in which an individual target is assessed. The contribution of this methodology is in providing a systematic and unbiased method for target assessment that can be scaled to any number of targets," said Stephen Frye, director of the Center for Integrative Chemical Biology and Drug Discovery at The University of North Carolina at Chapel Hill Eshelman School of Pharmacy.

"By combining biological validity and chemical tractability, this approach balances the debate in a helpful way and directs it toward practical drug discovery. At the very least, the risks in a portfolio of targets can be identified and targets ranked against them. In the end you make a decision based on a much more complex set of data than this, but as a first cut this provides an unbiased, comprehensive look," added Frye, who was previously worldwide VP of discovery medicinal chemistry at GlaxoSmithKline plc.

"If you were to give this to the heads of oncology at multiple pharmas, they would argue with the components or the weighing, but that wouldn't make this more right or more wrong. What this paper says is: here is a roadmap, a series of decisions that need to all be weighed," Friend told SciBX.

Expanding the dialog

A key next step for expanding the utility of the algorithm will be incorporating additional clinical data, said Workman. The next version, which will be released this summer, will include all of the genomic and expression data from the International Cancer Genome Consortium. This additional clinical data will expand the algorithm to help compare the level of biological validation of potential targets in addition to weighing likely druggability.

Friend noted that "there is a fair amount of data already included in the algorithm, but it's 1% of what could be in there."

Workman's team also plans to further refine the algorithm's ability to predict druggability.

Indeed, Frye said, "the druggability score is a fairly crude estimate for what is possible." For instance, he said, histone acetyltransferases come up as potentially druggable, but "people have been working on them for years and there still are not good inhibitors."

The algorithm also may miss druggable targets. The methyllysine-binding protein l(3)mbt-like 3 (L3MBTL3) was predicted to not be druggable. However, Frye and colleagues recently reported the first inhibitor of the target.2 Their molecule binds to a pocket that is only accessible when the target dimerizes and thus is not easily predicted by current algorithms.

"This is not target selection by computer," noted Workman. "The idea is to have all the data in front of you. You can look for those targets that are biologically really appealing, and then depending on your appetite for risk you can go for targets where there is chemical matter available or ones that are predicted to be druggable to varying degrees but have no chemical matter."

Workman said ICR is using the algorithm to prioritize its basket of identified potential drug targets, including those found via cancer genomics efforts and synthetic lethal screens in cancer cell lines.

The team looks for targets ranked toward the top by the algorithm and then chooses 5 or 10 targets to explore with more intensive wet biology experiments before making a final decision about which targets to pursue.

The method reported in the paper is unpatented and is freely available through the ICR's canSAR database.

Kotz, J. SciBX 6(6); doi:10.1038/scibx.2013.128
Published online Feb. 14, 2013

REFERENCES

1.   Patel, M.N. et al. Nat. Rev. Drug Discov.; published online Dec. 31, 2012; doi:10.1038/nrd3913
Contact: Bissan Al-Lazikani, The Institute of Cancer Research, London, U.K.
e-mail: bissan.al-lazikani@icr.ac.uk
Contact: Paul Workman, same affiliation as above
e-mail: paul.workman@icr.ac.uk

2.   James, L.I. et al. Nat. Chem. Biol.; published online Jan. 6, 2013; doi:10.1038/nchembio.1157

COMPANIES AND INSTITUTIONS MENTIONED

Cancer Research UK, London, U.K.

GlaxoSmithKline plc (LSE:GSK; NYSE:GSK), London, U.K.

The Institute of Cancer Research, Sutton, U.K.

Merck & Co. Inc. (NYSE:MRK), Whitehouse Station, N.J.

Sage Bionetworks, Seattle, Wash.

The University of North Carolina at Chapel Hill Eshelman School of Pharmacy, Chapel Hill, N.C.