Biovista’s drug safety search aids clinical outcomes.
By Kevin Davies
June 8, 2011 | ‘Biovista began as a hobby,” admits the company’s eloquent president, Aris Persidis. But after “dabbling” with the firm for a few years, Persidis incorporated the company in the United States in 2005. Headquartered in Charlottesville, Virginia, Biovista is becoming an important partner in the drug safety and repositioning arena.
Biovista started out in the mid 1990s as a business intelligence company, but Aris, a biochemist by training, and his cofounder (and brother) Andreas, a naval architect by training, would spend long hours debating whether new engineering approaches could answer some complicated biological questions, such as: Can one figure out a side effect before it happens, or deduce a mechanism in a disease when you don’t have a drug?
These two questions go hand in hand, says Persidis. “We can’t conceive contemplating efficacy and not simultaneously how it will impact safety and adverse events,” he says. “We quickly needed a technology to agnostically develop safety and efficacy simultaneously.”
Persidis set out to develop a technology called the Clinical Outcome Search Space (COSS), which features the ability to use information about mechanism to drive decisions around drug safety and activity. The brothers recruited scientists and physicians “that understood the question” and who also had expertise in coding, not the other way around.
Using text mining and resources from the National Library of Medicine, U.S. Food and Drug Administration (FDA) and other databases, Biovista began to tally as many unique indications and adverse events as it could find. “It was not easy, but we converged on about 23,000 unique indications and 6,000 adverse events,” says Persidis.
Persidis next asked: How do we interrogate 29,000 clinical outcomes, even in the absence of a drug? “We take 20,000 human targets and ask: Can we map the mechanism of action (MOA) of every target against the MOA of all 29,000 clinical outcomes? That was the genesis. We also overlaid the MOA of every one of 90,000 drugs and active compounds in the public domain against those 20,000 targets and 29,000 adverse events.”
The result, according to Persidis, is that COSS can match every human target against molecular information associated with every clinical outcome to generate a ranked list of 90,000 drug target hypotheses. “We think of COSS as a massive hypothesis-generation engine,” he says. “We take a package of targets, then we do a variety of things. First, we generate 90,000 drug hypotheses. COSS connects them mechanistically to the target package and ranks them in terms of degree of overlap of MOA versus that of the targets. This also works with combinations.”
Next, in order to match targets to indications, Biovista ranks all 23,000 indications in terms of their propensity of mechanistic overlap with the target. “We then replace an infinite set of possible hypotheses with a very tight set of a dozen to 50 per mechanism. The same with adverse events and drugs: we reduce an infinite number of combinations to a manageable few.”
Given that COSS can be used to evaluate safety signals in any drug class, it is not surprising the technology has attracted interest from the FDA. Biovista has worked with FDA informally for the past five years, collaborating to evaluate the safety of a number of selected drugs and drug classes using MOA.
“Larry [Lesko] was interested in our work attempting to map MOA,” explains Persidis. “[The Office of Clinical Pharmacology] set a series of tests for us, before the collaboration was more fully consummated and announced at the beginning of 2010. We had to work very hard to show the usefulness of this technology, which we did using drug scenarios supplied by FDA. We had to explain these adverse events after they happened, when the sponsor was unable to do so.”
Persidis declines to offer specific examples. “There’s a wall between what we deliver to FDA and how they communicate this to the trial sponsors,” he says. “So I can’t speak about drug identities or MOAs we’ve identified. But we look forward to the time when the standard operating procedure at FDA will involve a mechanistic analysis of all drug classes.”
One of Biovista’s established collaborations is with Pfizer, which has a large group in St. Louis dedicated to drug repositioning or “indication expansion” under the leadership of Don Frail (director for Pfizer’s St. Louis research site). Says Persidis: “How can we use mechanism to take Pfizer drugs and find something interesting to do with them that maybe Pfizer hasn’t considered? Can we go orthogonally and identify something genuinely novel and medically relevant? That’s the essence of the collaboration. The adverse event portion is a wonderful dimension and one we uniquely add.”
COSS ranks all adverse events in terms of their mechanistic propensity within the context of a therapeutic scenario. “It’s to help the sponsor to understand the inclusion/exclusion criteria, prior to the commitment of trials. It’s about choosing the subsets of populations,” says Persidis.
Rather than transfer the COSS technology to a user, Biovista uses the technology in house together with the partners’ data. “Our partners may have ‘omics and binding data that we use to refine what we’ll do, though we don’t necessarily need it.” COSS is used throughout the development cycle of a drug, even before it exists as an entity and one has a notion around a target. “It can navigate from disease to target to drug to adverse event,” says Persidis.
In time, Persidis hopes that COSS will bring about improved design of clinical trials. For example, patients susceptible to a known side effect of a given drug candidate, such as a kinase inhibitor, could be excluded from a new trial. Persidis says COSS is designed to complement long-standing tests to measure drug metabolism, such as the Long QT interval. “COSS looks after the rare adverse events, where it’s not simply a case of a cytochrome variant and drug metabolism but where the mechanism might increase prevalence of [a disease such as cancer].”
Biovista’s internal pipeline features two drugs for multiple sclerosis with novel MOAs, as well as candidates for epilepsy, glioblastoma and melanoma. Proof-of-concept trials are currently being designed.
“We need innovation, not just in biology but also business models. It’s mission critical for us to have a set of collaborations to demonstrate what we do,” says Persidis. “We don’t comfortably fit in the classic description of drug development, because we’re developing our own pipeline together with collaborating on the pipelines of our partners. That gives us insights and flexibility of business deployment that we think is very relevant.”
Persidis describes some promising results in pancreatic cancer. Using a simple Excel spreadsheet, Biovista plotted 30 validated targets against pancreatic cancer, then asked how many of those targets are hit by the major frontline therapies, such as 5-FU, Tarceva (Erlotinib) and Gemzar (Gemcidabine). The answer was only 9 out of 30.
“So current therapies hit only up to 1/3 of the reasonably well understood targets.” Biovista then took 90,000 drugs and asked: which ones are known to perturb the MOA represented by those 30 targets, but have never been used in pancreatic cancer? This highlighted a series of molecules that could potentially fill in some key gaps in the battle against pancreatic cancer.
Persidis says this is “a fully baked resource” that he hopes will be made publicly available under the Biovista Foundation that is being set up, “so patients, clinicians and physicians can map their work onto this resource.” He intends to keep some drugs for Biovista and make many others available to help the patient population. “This molecular epidemiology approach will be a significant new tool in the cancer fight and in other complex diseases, too.” •