Rethinking Clinical Trials with Faster AI-Driven Decision Making - with Shefali Kakar of Novartis

Exploration of how pharmaceutical companies are using computational modeling to design drugs at the molecular level before physical synthesis. This includes modeling every chemical change (carbon, hydrogen, nitrogen) to predict safety, pharmacokinetics, and efficacy across entire industry datasets, not just single programs.

• •Companies can now model molecular changes to predict safety, pharmacokinetics, and activity before synthesis
• •Third-party companies have emerged that perform in silico analysis for both small biotechs and large pharma
• •Protein engineering allows designing drugs that last a month in the body versus daily injections, improving market viability
• •Design decisions now incorporate competitive analysis against existing market drugs to ensure superiority

Discussion of how in silico clinical trials and probability of success models are reshaping investment decisions in pharma R&D portfolios. These models integrate market size, similar program data, and clinical trial predictions to inform capital allocation, though current implementations still have limitations.

• •In silico clinical trials can provide probability of success metrics before actual human trials begin
• •Current probability models are based on similar programs, market size, and historical data but lack cross-functional integration
• •Future refinement could lead to decisions about phase 3 trials based purely on probability modeling above certain thresholds
• •More sophisticated cross-functional integration into success modeling will improve accuracy over the next decade
• •Risk appetite will determine how organizations use these probability metrics in go/no-go decisions

Analysis of the critical gap in data access and formatting that currently limits AI capabilities in pharma. Companies are investing heavily in creating data lakes to aggregate information regardless of format, with particular emphasis on capturing failure data from discontinued programs and acquisitions.

• •Primary bottleneck is data access and format standardization, not AI capability itself
• •Companies are creating data lakes to first collect all data, then format it for analysis
• •Failure data from discontinued programs and acquisitions represents a largely untapped resource
• •Learning from failures is critical to avoid repeating mistakes, but this data is often lost in M&A activity
• •Next few years will focus on data infrastructure investment before full AI value realization

Shefali challenges the notion that 'fail fast' doesn't apply to pharma, arguing that with proper data access, early failure detection actually protects patients. Identifying drug failures in phase 1 (20 patients) versus phase 3 (thousands) represents a massive reduction in patient risk exposure.

• •Fail fast should apply in pharma when supported by comprehensive data to minimize patient exposure
• •Better data access enables identification of 'no-go' scenarios earlier in development
• •Phase 1 failures (20 patients) are vastly preferable to phase 3 failures (thousands of patients)
• •Early detection of ineffective or harmful drugs reduces both patient risk and R&D waste
• •Data-driven early termination decisions represent ethical patient protection, not just cost savings

Summary of three critical insights for life sciences leaders: the role of multi-factor modeling in pre-clinical decisions, AI's transformation of capital allocation through success probability metrics, and the structural changes needed around data access and learning from failure.

• •Multi-factor modeling simulates chemical interactions and patient outcomes before clinical trials begin
• •AI transforms capital allocation by gauging success probabilities across large R&D portfolios
• •Structural changes must prioritize access to siloed/unstructured data
• •Organizations must shift culture to learn from failure data, not just success stories
• •These capabilities are still scratching the surface of potential impact on drug development