Stop me if you’ve heard this before:
“Chat-GPT (and “Generative AI”) will take over the world.”
What does this really mean? At IDEA Pharma we have set out to help “demystify” the black box of artificial intelligence by laying out specific use cases where these tools can bring value to drug developers and, ultimately, patients.
But first, let’s set the stage:
Natural language processing (NLP) has the potential to revolutionize the drug discovery process for major pharmaceutical developers by synthesizing the vast troves of publicly available medical and scientific research available in the public record, forming the technological “backbone” for advances in Generative AI (such as OpenAI’s Chat-GPT, Google’s Bard, and other emerging solutions). With its ability to process enormous amounts of information, NLP offers a range of value propositions for pharmaceutical companies looking to streamline their drug discovery process, reduce costs, create novel value, and bring new treatments to market faster. Specifically, we at IDEA Pharma leverage these advances in technology to complement and accelerate the extraction of human-generated insight, accelerating the innovation flywheel by finding common ground between data analysis and the industry expertise needed to separate fact from fiction for our clients. Specifically, we at IDEA Pharma have identified several different ways in which NLP can be leveraged to help pharmaceutical developers bring additional value and optionality to their drug discovery process. Here are just a few of the specific use cases we’ve encountered through our journey to help more great medicines reach more patients:
1. New Target Discovery
“What ways can we target disease control in Multiple Myeloma’s explosive disease progression?”
NLP can help major pharmaceutical developers identify new targets for drug development by analysing vast amounts of scientific literature to identify gaps in knowledge and areas with high potential for new drug development. For example, by analysing the scientific literature on a specific disease, NLP can help experts identify key proteins or pathways involved across disease. This accelerated synthesis of vast troves of raw scientific, clinical, and commercial data augment the core human decision making process within pharmaceutical development, allowing our internal experts to expedite highlighting areas that have not yet been explored as potential drug targets. This can help major pharmaceutical developers prioritize their research and development efforts and increase the chances of success in bringing a new drug to market. It is important to note, however, that the process of drug development will always be human centered; molecules don’t change from discovery to launch, only the man-made decisions that drive what (and what not) to study. This is a key tenet of our philosophy at IDEA of asymmetric learning – human decisioning is a key, with AI as a tool, not a panacea.
(Side note: I’m going to take a moment now and further elucidate this specific case study; for subsequent questions, see how you can adapt the questions that can be posited to AI/ML/NLP tools to help improve the human decisioning behind drug development.)
AI tools can help generate hypotheses to explore for this specific question, further weighted and rank-ordered based on the business and clinical needs of the indication of interest. AI can help start the search, but really, this data needs to be further synthesized by industry experts to elucidate areas of potential value and the “lift” required to unlock that value. Does targeting disease control help enter a specific market of interest? What is a market of interest? Does it involve manufacturing capability, established brand identity, or genetic signatures of the population that would make this insight more valuable (such as the higher rates of gastric cancer in Asian markets)? AI can illuminate available data, but it up to the human element to link it back to “key strategic questions” – and defining what “key”, “strategic”, and “question” really means, unique to each organization, portfolio, and circumstance. While NLP can accelerate the initial data search, the idea of “streamlining the discovery process”, much like learning about a molecule that never changes from discovery to launch, is a human-centred workflow. Better decision is very different from making the decision. We at IDEA Pharma work closely with our clients on this exact challenge in regards to what AI is best suited to perform based on client needs; just because AI can “do” something doesn’t mean it “should”.
(Another side note: For additional context on AI’s accuracy and the fallacies of face-value belief, check out any number of articles exposing factually incorrect insights from Generative AI (here, here, and here) to illuminate what we mean about AI as a tool, and not a panacea for better decision making…AI can help the process of decision making, but asking it to make the actual decision without expert insight is a dangerous game to play, even more dangerous considering the stakes of healthcare).
With that being said, we will continue on with potential use cases we’ve encountered at IDEA through our deep analysis of how AI can provide additional value in the drug discovery process as a tool to be leveraged in alignment with better human decision making, versus a tool to make the decisions.
2. New Payload Discovery
“What novel mechanisms are being developed to generate further precision of DNA damaging agents in solid tumour oncology?”
NLP can also be used to identify new payloads or drug candidates by analysing information about molecular structure, pharmacology, and toxicity of various compounds. This information can help major pharmaceutical developers prioritize compounds for further development, streamlining the drug discovery process and reducing the time and cost required to bring a new drug to market. Again, please remember that this process needs to incorporate a view on what the market needs (as well as what it actually wants), which is a distinctly human-centred exercise…or, as we like to call it, part of IDEA’s 3-Legged Stool of Innovation). For example, is this simply about new payloads, or is it more valuable for our client to explore novel commercial/pricing models in the face of limited remaining patent life? A “new payload” could be a distinctly different molecular mechanism, or an established payload in a new indication of interest. The key bit here is “indication of interest” – is interest scientific, commercial, strategic, or a permutation of the three? Another human-focused decision point where AI can support, not replace. This segues neatly into our next use case:
3. Repurposing Previously Formulated Drugs
“What destinations of optionality exist throughout the lifecycle of ‘Drug X’ in neuropsychiatry?”
NLP can help major pharmaceutical developers repurpose previously formulated drugs for new indications and therapeutic purposes by analysing the literature to identify new use cases to explore for existing drugs. This process can result in faster and more cost-effective drug development, as the safety and efficacy of the drugs in question have already been established. Specifically, in this use case, AI can help Pharma developers generate and explore new, potentially valuable, biological and commercial hypotheses, enabling developers to think through what the next best experiment could and should be, while providing better visibility on key challenges and the evidence needed to address these challenges.
Here is the end of Part 1, keep an eye out for the next issue of IDEA Junkie, where I will be going into 5 more use cases that we’ve encountered, as well as what conclusions we draw from these specific use cases and what we at IDEA Pharma utilise in order to achieve our mission:
“To create path-to-market options that no-one else could, and inspire biopharma to develop and successfully launch more great medicines.”
Look forward to seeing you back for Part 2 where I will be discussing clinical trial designs, M&A, collaboration and many more.
In the meantime, reach out to us by email or LinkedIn if you have any thoughts or comments on this article.
