Valence’s co-founder and CEO, Daniel Cohen, spoke to Lu Rahman about the company’s plan to empower drug discovery scientists with AI-enabled drug design
LR: Outline of the company and its main initiatives
DC: Valence is an interdisciplinary team of scientists and engineers out of Mila, the world’s largest deep learning research institute. At Valence, we are focused on developing novel machine learning methods for molecular property prediction, generative chemistry, and multiparameter optimisation.
Our mission is to empower drug discovery scientists with the latest advances in AI-enabled drug design, providing them access to powerful technologies they wouldn’t otherwise have exposure to, so they can focus on the biology while we drive the chemistry into new and interesting directions.
We believe that by unifying best-in-class deep learning technologies with intuitive infrastructure to make these technologies more broadly accessible, we can accelerate the adoption of AI in drug discovery and ultimately support our partners in the design of higher quality drug candidates, faster than previously possible.
LR: How can your technology help the drug discovery sector and what savings or efficiencies can it help create?
DC: Only a tiny fraction of drug discovery organisations today are AI-enabled. There are a few reasons for this.
First, building top-tier AI capabilities isn’t a core competency for the majority of drug discovery organisations. It requires a dedicated team cross-trained in machine learning and the life sciences, and also an army of engineering support.
Second, the space is evolving at an extremely rapid pace. It can be very difficult to stay on top of the latest advances in the field if you’re not living in it every day and actively developing new AI methods internally.
Finally, scientists today are relying primarily on patchwork AI tools borrowed from other domains. These off-the-shelf solutions are rarely well-adapted to drug discovery and, more often than not, yield disappointing results.
Valence is addressing all these issues with the first plug-and-play AI infrastructure built specifically for drug design. Our platform unlocks novel therapeutic domains for our partners through the design of diverse chemical matter in untapped regions of chemical space, while also enabling rapid optimisation of these molecules against multiple potency, selectivity, safety, and ADME properties.
This enables meaningful time and cost savings in bringing first-in-class or best-in-class drug candidates into the clinic.
LR: You have formed significant collaborations recently – why is this and what will they help the business achieve?
DC: We recently announced collaborations with IRICoR, Canada’s largest academic drug discovery institute; Repare Therapeutics, a leader in precision oncology; Servier, an international pharmaceutical company; and Charles River, a global leader in preclinical research services for biotech and pharma.
With all these groups, we’re leveraging our platform for AI-enabled drug design to rapidly and cost-effectively unlock novel chemical space through the design of diverse chemical matter against previously intractable targets.
Alliances with partners who are trailblazers and drug discovery leaders are core to our mission. We’re providing our collaborators the freedom to innovate on the biology while we drive the chemistry into new and promising directions, under what is ultimately a highly collaborative, highly complementary partnering model.
LR: What is different about the Valence technology? Where do you feel it stands apart?
DC: The biggest limitation on deep learning in drug design is the lack of sufficiently large datasets to learn reliable structure-activity relationships for all the properties we need to predict when designing novel chemical matter. Whether it’s potency at a novel target, selectivity across a panel of off-targets, or ADMET properties in lead optimisation, standard (data hungry) deep learning tools fail under the data constraints of a typical discovery program, often when they’re needed most.
At Valence, our team has pioneered the application of few-shot learning in drug design, a technique that allows us to learn accurate predictive models from very small datasets, in turn unlocking a huge swath of previously intractable problems for deep learning in drug discovery.
On top of this, we’ve developed generative methods (for de novo molecular design) that allow us to enforce a high degree of quality and improve synthetic accessibility in the design process. AI-designed molecules are of little use if they can’t be readily synthesised for profiling in the relevant assay(s). Our platform ensures that this is always possible, addressing a major limitation of competing technologies.
Finally, our platform is powered by academia-leading research out of Mila, the world’s largest deep learning research institute, and we’re fortunate to count leading scientists, including deep learning pioneer Professor Yoshua Bengio, as close scientific advisors, ensuring we stay at the forefront of our field.
LR: Where do you see AI/ML technology market heading and how will it be of most benefit to the research sector?
DC: Today, we’re starting to see the impact. We’re seeing compounds enter the clinic from AI-first companies using these technologies to better interrogate biology and chemistry. We’re seeing major advances in complementary domains like protein folding, and we’ve seen meaningful commitments in AI/ML from most of the major pharmaceutical companies.
But we’re still in the earliest days of AI-enabled drug discovery. The next step for the space will be making these technologies more accessible to reap maximum benefit across the entire industry. This is our mission at Valence, and something we’re very excited to be contributing to.
LR: What’s next for the business?
DC: We believe that within the next decade, all drug candidates entering the clinic will have been discovered or designed with meaningful input from AI methods. We’re excited to be playing a key role in powering this shift.
We’ll be furthering our mission by partnering with additional biotech and pharmaceutical companies, and scaling our reach through strategic partnerships that allow us to bring our platform to a greater portion of the market.
We’ll also be expanding our presence in the Boston area as we grow our machine learning, engineering, chemistry, and business development teams.
