As the frenzy around Generative AI continues to heighten, each stakeholder in the system (ML engineers, scientists, and end users) continue to calibrate their viewpoints on the promise of the technology. Some are convinced that this is a fad that will pass. To be fair, it has happened before. Remember, Segway was supposed to change urban transportation but ended as a punchline. 

Some others continue to believe in the promise but at the same time refine and recalibrate their viewpoints with new developments. We fall into this category. As conjectures follow, I think Generative AI is a big-tent technology and multiple viewpoints can and will co-exist. We should of course feel free to change our views one way or another as we learn more. 

Before we dig deeper, I will narrow down the scope of this conversation. First- since my experiences and awareness are largely limited to the life sciences industry, I will stay in that lane. Second, I feel there are 4 different categories of expertise emerging around LLMs based on the scale of $$, Compute power, and, Skills required:

Just to set the context we are somewhere at the interface of group 2 and group 3. 

Now, let us dig deeper! 

The pharma and biotech industries have already started using artificial intelligence to improve efficiency, generate reports, and develop drugs. But like I already said, this is an uncharted territory, we are early adopters, and only time can validate the potential of the technology. Meanwhile, here are 3 ways 

Generative AI Can Transform Early-stage R&D

1. Improving Data Quality (Biocuration)
   
   I believe Generative AI is more than a buzzword and can create real tangible value for companies like Elucidata and in turn its customers who are discovering drugs. The key, however, is to define the scope of usage of the technology. It is perhaps an obvious thing to realize but worth repeating given that we all are collectively trying to educate ourselves as to what LLM means for drug discovery.
   As Ricks- CEO, Eli Lilly put it, ‘the technology could carry out the first mundane steps in tasks such as contract production or the rote parts of administrative work.’ Or that, ‘AI could also help automate repetitive business processes.’ In short, LLMs hold a huge promise. This is quite consistent with our own experience at Elucidata. For curating high-quality data, we have found LLMs to give us a huge advantage. We are experimenting with ChatGPT and prompt-based engineering to extract biomedical entities from publications. So far, we have achieved an accuracy and F1 score close to 83%, for sample-level disease extraction. You can read more about it here. 

2. Creation and Validation of De Novo Antibodies
   
   Another tried and tested use-case is the creation and validation of de novo antibodies. Generative AI is being used for performing virtual screening of large compound libraries, identifying potential ‘hit' molecules, or designing molecules de novo based on desired properties and structural features. Recently, Absci published their work on the generative design of antibodies to target three different disease-associated molecules (HER2 receptor, VEGF growth factor, and spike protein of SARS-CoV-2). They claim that these AI-designed antibodies can slash the drug discovery timeframes by upto 50%, while also increasing their probability of success in the clinic.
   Meier- Absci Senior VP and Chief AI Officer, in an interview stated, ‘The controllability of AI-designed antibodies will enable the creation of customized molecules for specific disease targets, leading to safer and more efficacious treatments than would be possible by traditional development approaches.’
   
   
3. Report Generation / Visualization
   
   One of the final steps for insight generation is to explore and visualize your data to gain insights and identify patterns. Traditional data visualization has been effective, simplifying intricate numerical data into easily digestible visuals. Yet, as data multiplies and complexity soars, conventional methods face challenges, particularly with massive datasets and real-time updates. Static visuals fall short in grasping personalized insights and dynamic data shifts.
   In my opinion, this is one of the more ‘urgent’ problems that can be addressed using Gen AI. The technology can be trained to dynamically transform vast amounts of complex, multi-dimensional, high-density data into interactive, and insightful representations to make it more understandable and accessible. Some have already started experimenting to analyze large datasets, detect trends, and create dynamic visual narratives that update in real-time, providing latest insights for informed decisions. How effectively though? I guess we’ll know soon enough. 

We will be discussing more about Generative AI in our annual event DataFAIR 2023 which will be held virtually on the 26th of October between 1PM and 4PM EST. Join us to hear from the best minds in the Biopharma/ Biotech space to explore the revolutionary potential of Generative AI in drug discovery. Click here!