It’s impossible to be online and not hear about ChatGPT, the controversial chatbot launched by OpenAI in November 2022. Here, DDW’s Diana Spencer takes a look at the ways ChatGPT and technology like it could change the future of drug discovery and development.
Chat Generative Pre-Trained Transformer (or ChatGPT) is a conversational chatbot, designed to interact as a human would. The creators OpenAI have made the prototype software available for free while it is in the testing phase and encourage users to share their feedback1.
On the company website, OpenAI says: “The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.”
And according to Wikipedia: “It has the ability to write and debug computer programs, to compose music, teleplays, fairy tales, and student essays; to answer test questions (sometimes, depending on the test, at a level above the average human test-taker); to write poetry and song lyrics; to emulate a Linux system; to simulate an entire chat room; to play games like tic-tac-toe; and to simulate an ATM.”
This versatility, paired also with a tendency to give incorrect answers, has led to concerns that it will no longer be possible to rely on online interactions being genuine and even less on the accuracy of information provided on the Internet.
So, how could software like this impact drug discovery, an industry that has so far embraced the possibilities of artificial intelligence (AI)?
In January 2023, Nature reported that ChatGPT was being listed as a contributing author on academic papers2, indicating that academic researchers are already beginning to use the bot in their studies. Many journal publishers argue that a chatbot can’t be listed as an author and are being forced to create policies about how it can be used, or not, in the research setting.
ChatGPT’s ability to write convincing essays is also an area for concern and has led to fears that standard forms of school and university assessments, like essays and dissertations, will become a thing of the past.
To combat academic plagiarism, student Edward Tian has created an app that can tell the difference between text written by a chatbot and that written by a human, GPTZero3, which could provide some hope for the future of written assessments.
Using the software to write essays might not actually give a student or researcher an advantage over their peers, however, as many have found the chatbot to be inaccurate in its responses when in-depth scientific knowledge is required.
Lorenzo Bombardelli, a Business Developer and Scientist based in Amsterdam, posted on LinkedIn: “I was curious to see whether #AI monster #chatGPT could really replace 100s of study hours and years of work. Given that some universities banned it already, worrying that anyone could suddenty pass any test, I tried asking a hard scientific question requiring specific knowledge in the field of #genetics.”
In response to a request to “Suggest a light-inducible recombinase”, ChatGPT provided a well-written and convincing response that was entirely wrong, proving that AI like this has a long way to go before it can match or mimic a well-read researcher.
One way that ChatGPT has been tested in relation to the drug discovery sector is in regards to regulatory affairs.
The website Pharmavibes3 suggests that it could be a useful source of information about regulations in different countries and in helping companies to understand those regulations in relation to their own product. It also suggests companies could use ChatGPT when putting together applications to regulatory agencies.
It asked ChatGPT various very detailed questions related to regulations for medicines, and shared the detailed responses, though cautioned that any information provided by ChatGPT should be checked before included in applications.
Overall, the author seemed impressed and concluded: “It seems as though the responses are at least as good as if not better than the responses that one might receive from a human being…If chatGPT is a glimmer of the added value that AI might bring to regulatory affairs, it looks very promising and I look forward to trying out future iterations.”
In a recent study, Gaurav Sharma and Abhishek Thakurb from the Department of Chemistry, Michigan State University, USA, tested ChatGPT’s abilities in relation to the process of drug discovery, in particular, computational chemistry4.
In response to many of their requests, the researchers found ChatGPT accurate and helpful. It was able to compute the compound multiplicity, generate the input file for gaussian software and find the required PDB (protein data bank) files. It was also useful for the literature search, checking for plagiarism and writing basic code.
However, they found shortcomings in relation to answering complex questions, and providing the FASTA sequence and ADMET properties.
Sharma and Thakurb suggest ChatGPT could play an important role in identifying and validating new drug targets, designing new drugs, optimising drug properties, assessing toxicity and generating drug-related reports and papers.
They conclude: “It’s important to note that ChatGPT is just one tool among many that are used in drug discovery, and it is not a substitute for experimental validation and clinical trials. However, by providing a cost-effective and efficient way to process large amounts of data and generate new knowledge, ChatGPT can assist researchers in making more informed decisions and accelerate the drug discovery process.”
Protein language models
While ChatGPT may have its limitations, researchers are already looking into the possibilities of the technology behind it – natural language processing5. The approach requires teaching an AI programme to analyse and synthesise proteins, in the same way that ChatGPT can be taught to recognise and respond realistically to language requests.
Karen Hao explains in The Wall Street Journal: “The models encode what might be called the grammar of proteins – the rules that govern which amino acid combinations yield specific therapeutic properties – to predict the sequences of letters that could become the basis of new drug molecules. As a result, the time required for the early stages of drug discovery could shrink from years to months.”
Various established and start-up companies are now using this approach to enhance known molecules, but it is hoped that when refined, the technology could identify ways to target areas that have so far been deemed undruggable.
ChatGPT has huge potential and is certainly far advanced when compared to other chatbots, but it is early days, and it will probably be a long time, if ever, before we can rely on an AI chatbot completely.
Acknowledging the current software’s shortcomings, in December CEO Sam Altman, CEO of OpenAI, tweeted: “It’s a mistake to be relying on it for anything important right now. It’s a preview of progress; we have lots of work to do on robustness and truthfulness.”
For now, it is perhaps better used in fun experiments than in real world applications, but as a proof of concept, ChatGPT could be a hugely influential step towards the future of AI in drug discovery.