According to Forbes, developing a new drug costs more than $2 billion and takes over 10 years, with 90% of drug candidates failing in clinical trials. Compugen, founded in 1993, has evolved from providing computational hardware to becoming a clinical-stage innovator with four clinical programs and partnerships with major pharmaceutical companies including Bayer, Bristol Myers Squibb, Gilead, and AstraZeneca. Executive Chairman Anat Cohen-Dayag emphasizes that success requires combining AI capabilities with deep drug development expertise, noting that 2025 could mark a tipping point for AI in life sciences. The broader industry is seeing massive investment, with AI drug discovery accounting for most of the $3.8 billion in AI funding for drug R&D in 2024, and CB Insights projects AI could generate over $350 billion annually for pharma. This momentum represents both unprecedented opportunity and potential bubble territory.
Sponsored content — provided for informational and promotional purposes.
The Computational-Clinical Divide
What most AI-pharma startups won’t tell you is that computational predictions represent only the very beginning of a long, expensive journey. While AI can rapidly identify potential drug targets from genomic data, the real challenge begins when these candidates enter the drug development lifecycle. Biological systems are notoriously complex and context-dependent—what works in a computational model often fails in living organisms due to unforeseen interactions, toxicity issues, or metabolic challenges. Compugen’s 30-year evolution from computational service provider to clinical-stage company demonstrates how difficult this transition truly is. Many current AI startups lack this depth of institutional knowledge and may underestimate the biological validation required beyond algorithmic success.
The Data Quality Conundrum
The pharmaceutical industry’s enthusiasm for AI often overlooks a fundamental limitation: garbage in, garbage out. While companies like Nvidia partner with Eli Lilly to build “AI factories” with massive computing power, the quality and relevance of training data ultimately determine model performance. Pharmaceutical data is often fragmented across proprietary systems, collected using different protocols, and plagued by publication bias where negative results go unreported. Furthermore, biological data has inherent noise and variability that can mislead even sophisticated algorithms. Compugen’s approach of “feeding back learning from failures” into their platform acknowledges this reality—something newer entrants may discover only after burning through significant funding.
The Unseen Regulatory Challenges
As AI-generated drug candidates advance toward clinical trials, they face unprecedented regulatory scrutiny. The FDA and other global health authorities are still developing frameworks for evaluating AI-derived therapeutics, creating uncertainty about approval pathways. Regulators will demand extensive validation of both the AI models and the resulting compounds, requiring transparency about training data, algorithm biases, and decision-making processes. Companies that cannot explain exactly how their AI arrived at a particular drug candidate may face significant delays or rejections. This represents a particular challenge for “black box” AI systems where the reasoning process isn’t easily interpretable by human experts.
Separating Substance From Hype
The current investment frenzy around AI in pharma shows classic bubble characteristics. With Isomorphic Labs raising $600 million in a single funding round and numerous startups describing themselves as AI developers, the field is becoming crowded with companies that may not have sustainable business models. The fundamental economics of drug development haven’t changed—success still requires navigating complex biology, rigorous clinical trials, and unpredictable regulatory environments. Investors drawn by the promise of AI may underestimate how long it takes to generate revenue, with many companies years away from having products that can even enter clinical testing, let alone reach market.
Why Domain Expertise Still Matters
Compugen’s transition under Cohen-Dayag’s leadership highlights a critical insight often missing from AI discussions: technology cannot replace deep domain expertise. Understanding “why and how and what would make a good drug target” requires biological intuition honed through years of research and practical experience. The most successful implementations of AI in drug discovery will likely come from companies that integrate computational approaches with seasoned pharmaceutical veterans who understand the nuances of disease biology, clinical trial design, and regulatory strategy. As Compugen’s journey shows, this combination of computational power and human intelligence represents the most promising path forward, but it’s also the most difficult to achieve.
The coming years will separate the genuinely transformative AI-pharma companies from those simply riding the investment wave. Success will require not just technological innovation but also biological wisdom, regulatory savvy, and the patience to navigate the decade-long journey from algorithm to approved therapy.
