A simplistic neural network diagram [Adobe Stock]
There are two courses of action pharma companies could take in such a reality. “One of them is increasing headcount and all of the activities in that seven to 10 year span from post-discovery up through regulatory and commercial approval,” Latshaw said. But the talent pool is likely not big enough to double the headcount of all pharma companies in development. “The only option that’s left is technology,” Latshaw said. “How do you harness technology to manage that process and that many extra molecules more efficiently?”
Separating AI fact from fiction
While the potential of using AI-based tools such as AlphaFold 3, RoseTTAFold All-Atom, machine learning for target identification and so forth, is clear, the path to realizing promised benefits is less straightforward. Pharma companies seeking to decode how AI can help improve the quality and quantity of their pipelines must also navigate a landscape rife with hype and misconception.
Dave Latshaw, Ph.D.
For all the buzz — and often confusion — surrounding AI’s potential in drug development, Latshaw injects a dose of reality. “There’s oftentimes a disconnect between what companies are doing and what they say they’re doing,” he cautioned. While not necessarily intentional deception, Latshaw observes a tendency to fit narratives around popular trends, rather than letting the work speak for itself.
He cites the example of some so-called “AI-discovered” molecules, which often turn out to be the product of traditional research methods with AI applied post-hoc for validation. This “varnish” on the narrative, as Latshaw described it, can distort public perception and create unrealistic expectations.
The true measure of AI’s impact lies not in marketing buzzwords, but in tangible progress. “My general thesis here,” he explained, “is that we will be successful in using AI to significantly increase productivity in discovery.” This optimism is grounded in the potential of tools like BioPhy’s BioLogicAI, which can forecast clinical trial endpoints with significant accuracy.
But achieving organization-wide success hinges on a crucial factor: the ability to manage the subsequent stages of drug development. “If you just have a giant bottleneck in development, you’re not going to see any of [the benefits of AI in discovery]. You’ll see basically the same level of productivity.”
Dimensional compression
Latshaw’s own interest in using deploying data science tools grew when trying to make sense of a data deluge working after arriving at a Big Pharma company in 2014 as a scientist in technical operations. Confronted with a dizzying array of 60 charts tracking each batch of antibody production, he sought ways to collapse this complexity while also uncovering the hidden relationships that traditional monitoring missed? Thus began his foray into dimensional compression, a technique that would transform not just data visualization, but influence R&D efforts at scale.
When presenting the idea to colleagues, some were initially skeptical. “What are you doing? We’ve done it this way for a long time,” they’d say. But he succeeded in deploying the pilot which “no surprise, uncovered that the univariate space wasn’t sufficient to actually understand the process,” Latshaw recalled.
By applying dimensional compression, Latshaw’s system transformed the way J&J monitored drug production. Instead of tracking 60 individual data points, the system extracted key features and combined them into a smaller set of latent variables. This provided a much clearer view of the manufacturing process.
Over time, Latshaw’s initial project grew into a large-scale ML R&D program that was implemented across multiple product lines and geographical regions. The project would go on to receive external recognition from the World Economic Forum and McKinsey & Company.
BioPhy’s two-pronged approach to confronting data heterogeneity
Latshaw’s current venture, BioPhy, has two distinct AI platforms, each designed to address a critical aspect of the drug development pipeline. BioLogicAI uses patent-pending technology to shed light on molecular interactions, analyzing the structural and chemical properties of the drugs themselves, structural properties of the biological entities, and the structural relationships between all of those biological things. This approach, reportedly boasting more than 80% accuracy in forecasting clinical trial endpoints, enables BioPhy to predict drug efficacy and potential toxicity. BioLogicAI guides assets through each clinical trial phase by continually assessing the likelihood of success based on multivariate inputs including mechanism of action, trial design, personnel, and operations.
Conversely, BioPhyRx is a proprietary generative AI platform developed by BioPhy, specifically designed for life sciences and drug development applications. It focuses on enhancing productivity across regulatory, quality, clinical, and operational workflows in the pharmaceutical industry. BioPhyRx provides on-demand scientific and regulatory guidance, as well as automating key processes such as standard operating procedure (SOP) gap analysis. This allows experts to accelerate core drug development functions, from rapidly accessing and interpreting regulatory standards to generating draft submissions tailored to regional requirements in real time.
Both platforms work in parallel to identify the highest potential candidates while dynamically generating conclusions to accelerate them efficiently through each development stage. This end-to-end approach tackles roadblocks from pre-clinical through approvals and beyond.
The concerted approach between BioLogicAI and BioPhyRx exemplifies Latshaw’s vision for a comprehensive, AI-driven transformation of the pharmaceutical industry. “The question isn’t whether AI will transform drug development,” Latshaw asserts, “but whether pharmaceutical companies will transform themselves to fully harness its potential.”
Filed Under: clinical trials, Drug Discovery, Drug Discovery and Development, machine learning and AI
