2nd February 2024 by Pratik Mitra | Pharmaceutical
The healthcare sector has already witnessed a significant impact from artificial intelligence (AI), as evidenced by the FDA's approval of over 520 medical devices utilizing AI. Traditional AI systems, often rule-based or statistical, operate within predefined parameters or make predictions based on historical data. However, their adaptability to new situations or learning from fresh data can be constrained.
In contrast, Generative AI (GenAI), a more recent AI subset, boasts enhanced flexibility and adaptability compared to traditional systems. GenAI systems can learn from data and generate new content, spanning text, images, or code. Trained on vast datasets, these models can generate novel data in the same format as their training data, be it text, images, or molecules. Notable GenAI models include Large Language Models (LLM) and Generative Pre-Trained Models (GPT).
The transition from AI to GenAI in Digital Health is still in its nascent phase, marked by the incorporation of new ventures and the adaptation of existing ones to exploit newfound advantages. Delineating ventures using AI versus GenAI proves challenging, warranting a comprehensive analysis of both technologies. The integration of GenAI holds transformative potential for healthcare delivery, offering the prospect of alleviating strain on currently overwhelmed healthcare systems.
Reduced Drug Development Costs: GenAI systems find application in identifying new drug targets and designing drugs by simulating molecular interactions and predicting potential candidates. An example is Insilico Medicine, which utilized GenAI to discover a treatment for idiopathic pulmonary fibrosis, a rare respiratory disease. This approach holds promise for reducing the costs associated with drug development.
Increased Healthcare Efficiency: GenAI facilitates task automation traditionally handled by humans, encompassing documentation, data analysis, and medical image interpretation. Athelas, based in California, utilizes GenAI to automate clinic documentation, covering clinical notes, patient reminders, billing forms, and more. Human-in-the-loop (HITL) models ensure healthcare professionals retain control over the generated text for each patient interaction.
Enhanced Accuracy and Precision in Diagnosis and Treatment: GenAI systems leverage data to discern patterns that might elude human observation, leading to heightened accuracy and precision in tasks like diagnosis, treatment planning, and drug discovery. Notably, Atomwise has developed a GenAI model claiming to diagnose skin cancer with dermatologist-level accuracy through the analysis of medical images.
Data Availability and Quality: Generative AI models demand substantial amounts of data for training, and this data must meet stringent quality standards. In healthcare, acquiring such data proves challenging due to the sensitivity and confidentiality of patient information.
Data Ownership: Establishing clear and mutually beneficial data ownership agreements is crucial. These agreements should respect the interests and expectations of all parties involved, including data subjects and stakeholders.
Regulatory Compliance: GenAI models used in healthcare must adhere to diverse regulations, especially those governing patient privacy and data security. Negotiating this regulatory landscape can be intricate and time-consuming, compounded by the evolving and region-dependent nature of rules and regulations for GenAI usage.
Bias in GenAI Models: GenAI models may inadvertently perpetuate biases if trained on unrepresentative datasets. This raises concerns about the accuracy and fairness of results generated by these models.
Patient Privacy and Security: The ethical considerations surrounding GenAI usage in healthcare include the legal implications of data ownership and the potential impact on patient privacy. The sensitive nature of patient data also makes GenAI models susceptible to cyberattacks, posing a significant security concern within the healthcare domain.
DRUG TESTING AI: The recent surge in generative AI has sparked both fascination and apprehension regarding its application in drug discovery and research and development. The current market value of AI in drug discovery is estimated at approximately $1.4 billion, with projections indicating a substantial rise to about $3.7 billion by 2027.1 AI-driven startups are pledging "unprecedented levels of efficacy and precision,"2 boasting of "revolutionary drug discovery engines,"3 and claiming to "radically improve the timely creation of high-quality drug candidates."4 Given the substantial time and financial commitments associated with conventional drug discovery and development, the anticipation for expedited and more successful drug development has generated significant excitement in the biopharmaceutical industry.
However, amid the optimism, there are notable barriers—both technical and ethical—that continue to shape the landscape of AI in the biopharmaceutical sector. Despite mounting evidence suggesting that AI has the potential to reshape the drug discovery and research and development processes, hurdles persist, necessitating a careful examination of both the promises and challenges associated with the integration of AI in this critical industry.
Considerable enthusiasm surrounds the potential of AI to streamline costs and timelines in candidate selection and preclinical research phases. The recent milestone achievement with INS018-055, the first AI-generated drug entering Phase 2 trials, suggests that AI may indeed deliver on these objectives. Developed by Insilico Medicine, INS018-055, a small molecule inhibitor designed for idiopathic pulmonary fibrosis, utilized two AI platforms—PandaOmics for target identification and Chemistry42 for molecule generation. These platforms significantly condensed the time for preclinical candidate selection from the conventional three to six years to a mere 18 months, with a cost of approximately $1.8 million. This remarkable success isn't isolated, as similar trends are evident across the industry. Exscienta's three drug candidates in clinical trials were AI-designed in under 12 months, and Benevolent AI's BEN-8744 was proposed for clinical studies within two years of program initiation. In academic settings, machine learning (ML) accelerated the search for candidate antimicrobials, exemplified by Dr. Gary Liu's team, which used a message-passing neural network to swiftly identify a narrow-spectrum antibiotic against A. baumannii, completing prediction and molecule prioritization within hours. The question arises: can AI truly "rewrite the playbook" for drug discovery? Arguably, in early drug discovery, AI is not rewriting the playbook but expanding its scope and accelerating its progression. While AI follows similar processes as conventional drug discovery—selecting a target based on scientific data and generating and screening compounds for target selectivity—the computational power supporting AI allows these processes to occur at an accelerated rate and, in some cases, simultaneously. This accelerated pace holds potential benefits for patients with high unmet needs, especially those with rare diseases lacking approved targeted treatments, and may contribute to addressing the antimicrobial resistance crisis by facilitating the rapid development of novel drugs, allowing researchers to stay ahead of the resistance curve.
The world's leading pharmaceutical companies have been exploring the applications of artificial intelligence (AI) for various purposes, such as drug discovery, patient identification for clinical trials, and understanding biological pathways. While these applications have seen some success, the actual feasibility has often fallen short of the grand predictions of a fully computer-driven future. Companies like Pfizer acknowledge that in reality, employing AI is more akin to training a super-intern than building a mechanical overlord.
However, major pharmaceutical players, including Pfizer, Merck, AstraZeneca, Sanofi, Amgen, and others, continue to forge new partnerships with AI-driven companies, as evidenced by numerous deals inked in the first half of 2022 alone. Notably, this trend extends beyond Big Pharma, with AI developers also engaging in consolidation. For instance, Verseon, a clinical-stage company with its AI drug discovery platform, acquired AI dataset expert Edammo in October to strengthen its capabilities.
The truth is that pharmaceutical investment in AI has surged significantly, growing from less than $1 billion in 2015 to over $7 billion in 2021, according to a report from life science. While pharmaceutical companies are actively embracing AI, the question remains: will AI be a fundamental pillar of the industry's future or is it an overhyped fad? The answer is complex, and elements of truth exist on both sides of the debate.
In conclusion, the pharmaceutical industry's increasing investment in artificial intelligence (AI) reflects a growing recognition of the technology's potential to transform various aspects of drug discovery, clinical trials, and understanding biological pathways. The industry's engagement in partnerships with AI-driven companies and the surge in investment, reaching over $7 billion in 2021, signify a commitment to integrating AI into pharmaceutical processes. While AI has shown promising results, especially in cases like the successful development of INS018-055 and the acceleration of drug discovery timelines, there is acknowledgment within the industry that the reality of AI implementation is more nuanced than some earlier visions. Analogies like "training a super-intern" highlight the practical challenges and complexities involved in incorporating AI into pharmaceutical workflows.The ongoing deals and collaborations suggest a continued belief in the potential of AI to enhance efficiency, reduce costs, and address critical challenges in drug development. Whether AI becomes a definitive pillar of the pharmaceutical industry's future or is deemed an overhyped fad will likely depend on the industry's ability to navigate and overcome the technical and ethical challenges associated with AI implementation. As AI technologies evolve and demonstrate their value, they may indeed play a transformative role in shaping the future of pharmaceutical research and development.