On September 5, 2014, GoogleNet’s advanced learning program achieved human-level performance in the ImageNet competition. The team’s performance in the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) that year was detailed in a blog post by Google Research. Insilico Medicine, established in 2014, focuses on leveraging deep learning for aging analysis and drug discovery. The organization will commemorate its tenth anniversary in 2024. In 2016, I published my second conceptual AI papers for AIDD. Generative Tensorial Reinforcement Learning (GENTRL) was developed in 2017 and officially released in 2019, with validation extending to mice. This validation took place in 2018. The document, spanning a year, not only generated significant interest in relational chemistry but also sparked debates among traditional computing physics firms. Despite a significant conflict of interest, I have closely observed the evolution of AI-driven medicine from its inception. In two forthcoming posts focusing on business and technology, I will make several predictions for 2024 based on current technological and business trends, as well as my field experience.
Consolidation in the AIDD Sector
The AIDD sector is expected to witness continued consolidation in 2024, following the trend from 2023. Established companies predating the deep learning revolution in 2014 and first-generation AIDD firms founded between 2014 and 2020 are likely to face retention challenges. Discerning investors may grow increasingly frustrated with the industry upon scrutinizing pipelines, revenue streams, partnerships, and development rates. Notably, no platform-level AIDD-pharma collaboration initiated between 2016 and the present has progressed beyond Phase I. Many high-profile discovery partnerships announced in 2019–2020 have yet to yield a preliminary candidate (PCC). This highlights the struggle of AIDD firms to deliver on promised advancements within the anticipated timeframe. In contrast, pharmaceutical companies, equipped with in-house AI researchers and access to proven AI programs, tend to engage with AIDD firms based on relationships rather than outcomes. Some AIDD entities are downsizing AI teams and refocusing on internal pipelines, often targeting dated or partially monetized objectives and early-stage Phase II programs.
Expect an increase in cost-effective acquisitions through acquihires or substantial downsizing for AIDD firms unable to progress human clinical trials with AI-discovered or IoT-designed therapeutics for novel or challenging targets.
Emergence of New AI Drug Discovery Startups
Despite limited understanding of the biotechnology sector, numerous AI innovators are poised to delve into drug discovery in 2024. While some may identify niche opportunities or stumble upon success without a clear niche, others, like ByteDance, may venture into this domain. It is crucial for entrepreneurs and investors to recognize the relatively small market size of pharmaceutical AI software and the necessity of developing proprietary medications for business success. The upcoming year is likely to witness a surge in such startups, potentially leading to a pool of skilled tech-bio professionals available for hire or entrepreneurial endeavors. However, a significant portion of these startups may face challenges, pivoting, or failing.
Evolution of System Partnerships into Asset or Software Licensing Models
The allure of end-to-end conceptual AI-enabled drug discovery lies in speed, cost-effectiveness, quality of protein and target hypotheses, innovation, and increased success rates. Many AIDD-pharma partnerships falter not due to technical shortcomings but rather due to coordination delays and friction at the initial stages. Consequently, AIDD firms with proven AI platforms have begun developing their pipelines, focusing on select applications to overcome these challenges. As these firms demonstrate AI’s potential, they are scaling up target exploration and producing high-quality molecules with desired attributes.
By prioritizing first-in-class (FIC) candidates for high-novelty targets or best-in-class (BIC) targets for moderately novel objectives, Insilico delivered 17 preclinical candidates in under three years. Pharmaceutical companies are adept at advancing these candidates through clinical trials and making them accessible to consumers. With several such successful collaborations underway, pharmaceutical firms are increasingly licensing products from AIDD companies, a trend likely to gain momentum in 2024.
NVIDIA’s Role in AIDD Software Development
NVIDIA’s GPUs have been instrumental in enabling advancements in deep learning. Since its establishment in 2013–2014, NVIDIA’s CEO Jensen Huang has prioritized dominating deep learning as a transformative technology. In contrast, other companies, like ATI (later acquired by AMD), initially overlooked this trend. Jensen’s strategic shift towards healthcare around 2015, following engagements with pharmaceuticals on deep learning’s potential, has proven highly successful. With partnerships like the one forged with Genentech, NVIDIA’s program is poised for further sophistication, albeit with validation levels below traditional therapeutic or computational chemistry standards.
An anticipated significant medical system launch by NVIDIA in 2024 may reshape the landscape for new AIDD startups. By offering resources for drug discovery via cloud platforms and local GPU clusters, NVIDIA aims to empower industry players while potentially stifling differentiation among emerging startups.
Anticipated Pharmaceutical Collaborations by Prominent Labs
Google DeepMind leads the AI landscape, consistently pushing boundaries with innovations. The introduction of AlphaFold in 2018 marked DeepMind’s foray into biotechnology, culminating in the groundbreaking AlphaFold2’s success at the 2022 protein folding CASP competition. While AlphaFold is widely adopted for research and publication purposes, no clinical trials have originated from the platform to date. DeepMind’s merger with Isomorphic Labs, spearheaded by Demis Hassabis, signals a strategic move towards advancing AI capabilities in drug discovery. Expectations are high for Isomorphic to leverage small-molecule science for challenging drug targets.
Noteworthy collaborations with Novartis and Eli Lilly, announced by Isomorphic Labs, hint at the sector’s support for their innovative approaches. As more alliances are anticipated in 2024, the industry is poised for transformative advancements driven by cutting-edge AI technologies.
Rise of Large Language Models (LLMs) in Pharmaceuticals
The proliferation of large language models (LLMs) gained momentum with ChatGPT’s launch in late 2022. In 2023, major pharmaceutical companies embraced LLMs for diverse applications, with Sanofi’s CEO endorsing an AI-centric strategy. Microsoft’s platform is poised to drive a surge in LLM adoption across the pharmaceutical and biotech sectors in 2024. While some LLM startups may outperform benchmarks, their integration into the biotech industry faces compliance and legal challenges. Microsoft’s Azure Cloud simplifies the deployment of advanced AI architectures and OpenAI models, positioning it as a global leader in facilitating LLM development.
In summary, 2024 promises significant advancements in relational AI and AI-powered drug discovery. However, the AIDD sector is expected to witness ongoing consolidation, paving the way for new startups, evolving partnership models, and transformative collaborations in the pharmaceutical landscape.
