Concurrent Pharmaceuticals, an early-stage drug discovery firm based in Fort Washington, Pa., last week announced that it has entered into a collaboration with Intel to develop machine-learning tools for computational drug discovery. In addition, Intel’s strategic investment arm, Intel Capital, joined the company’s previous investors — Prospect Venture Partners, Venrock Associates, and New Enterprise Associates — in a Series B preferred stock financing round totaling $15 million.
Concurrent is only the second life sciences firm that Intel Capital has invested in, out of a portfolio that includes over 400 early-stage technology companies. Intel Capital spokeswoman Laura Anderson said the company began looking into the life sciences sector about a year and a half ago after realizing that “a lot of drug research companies run very computationally intensive algorithms, and they’re running a lot of data through their processes.” Investments in companies like Concurrent are in line with the investment team’s overall strategy to support “efforts that might help drive demand for Intel products,” she said.
For Concurrent, the new funding is only one component of its relationship with Intel, which also involves a collaboration around the OpenML suite of machine-learning algorithms that the chip manufacturer released publicly in December [BioInform 01-05-04]. “By interacting with Intel we can speed up the testing and the investigation of different approaches in machine learning,” said Jean-Pierre Wery, vice president of computational drug discovery. “We are going to test some of the machine-learning tools that they have created and see how they perform on life science problems.”
Concurrent already relies on machine-learning to optimize its small-molecule discovery software, Wery said. The company’s platform is built around a set of algorithms it licensed from Harvard called SMoG (small molecule growth algorithm) and CombiSMoG (combinatorial small molecule growth algorithm). But after running the algorithms in the production setting that it put in place in the first quarter of 2003, “we realized that it would not afford us the accuracy that the chemists wanted to see to embrace these technologies, so we had to modify it substantially.”
Wery built upon the Harvard algorithms, which relied solely on the physical chemistry behind the small-molecule binding mechanism, by “taking advantage of the mass of data that we and other people are creating in structural biology … So every piece of data we create, or every piece of data that becomes available publicly, we can use in retraining to increase the accuracy of the algorithm.”
The approach fosters what Wery described as a “learning culture” at the company: “Before we decide to make a molecule, we know why we are making the molecule — it’s based on a belief that the molecule is going to be active, and have the right properties. So if we make it and it behaves as predicted, that’s fine. But if it doesn’t behave as predicted, then we can right away put this piece of data back into the training set and modify the algorithm, so very quickly we can reach a very high level of accuracy.”
The company’s computational suite, which it calls the Concentric technology platform, includes four components: Explorer Rx, a 3D drug-design engine that assembles compounds in silico from a set of compound fragment libraries; ConTour, a set of 2D and 3D database search engines for in silico screening; Gauntlet, a set of “drug-likeness filters” to predict ADMET properties; and Conductor, a PC-based interface for the company’s chemists and biologists.
On the surface, Concurrent’s software offers many of the same features as many off-the shelf computational chemistry tools, but the difference, according to Wery, is that the customers for commercial tools “are typically the computational groups in companies, so that doesn’t address the silo between the computational group and the medicinal chemistry group. Our software is really built to be used by the chemists.”
Rich Baxter, chief business officer of Concurrent, said that one of the company’s key goals was to improve the interaction between the chemists and computational scientists. Traditionally, he said, “rather than being integrated and working together in teams, it’s basically been that the computational people give the chemists a design and the chemists say, ‘Do this, do that,’ and throw it back over the wall.” Wery cited this disconnect as the reason that “computational tools have not made a big impact so far” in drug discovery.”
The advantage of a home-grown system, Wery said, is that any changes that a chemist requests in the software can be made “almost on the spot,” making the medicinal chemists “much more comfortable with using these tools day in and day out.”
According to Baxter, the strategy is already bearing fruit. The company expects a compound from its lead program — a renin inhibitor to treat hypertension — to be in clinical testing by early 2005, he said.
Currently, the company is focusing its computational drug design efforts on well-characterized targets in the areas of cardiovascular disease, central nervous system disorders, and metabolic disease, and therefore does not rely heavily on bioinformatics methods to discover novel targets, Wery said. In the future, however, “we might expand and pick more speculative targets and then engage more bioinformatics tools,” he said. “We obviously have a lot of the machine-learning framework and so it would be easy for us to go into using bioinformatics tools, but we haven’t reached that aspect of the program yet.”