Test and Prove the Value of AI in Drug Discovery

The rise of artificial intelligence (AI) has garnered significant attention in the field of drug discovery, fuelled by substantial investments from numerous companies and many claims about its positive impact. As AI becomes more integrated into research and development, an increasing number of professionals are asking the same essential question: “Am I getting value from AI?”

In this article, we focus on the early stages of integrating AI into your drug discovery projects, specifically when you’re evaluating a new AI application before it becomes a standardized part of your processes. While we’ll discuss methods for proving AI’s ongoing value in a future blog, the goal here is to help you assess its initial impact.

Defining value is the best place to start

Before diving into the specifics of testing AI’s value, the first step is to define what “value” means for your organisation. In drug discovery, we often rely on traditional metrics, such as model statistics, to gauge the success of AI methods. Strong statistical performance, including accuracy and reliability, is undeniably important, as it builds confidence in the models you develop. However, focusing solely on statistical outcomes can obscure the broader value AI offers. Model statistics can tell us that a method is performing well, but they don’t necessarily reflect AI’s true impact on your research process or the efficiency it brings.

Let’s clarify with an example: If your drug discovery pipeline is already optimised, you have quality data on your drug targets, can move swift swiftly from concept to candidate, with simple, scalable syntheses and make well-informed decisions on compound and experiment prioritisation—AI may not be necessary for you. However, if you feel there’s room for improvement, this is where AI can make a meaningful difference. It’s not about wholesale changes; rather, AI can help implement small, incremental improvements that ultimately lead to significant gains in efficiency and decision-making.

This brings us to a critical distinction: AI is fundamentally an enabler. It’s a tool that enhances your existing processes and doesn’t necessarily result in wholesale change, yet defining value can still be notoriously difficult.

Across any given drug discovery project, AI can add value in multiple areas: reducing the number of compounds synthesised, minimising the volume of experiments run, reducing wastage, optimising reaction parameters…the list can go on.

When aligning on value, there might not be an easy answer or there may be multiple answers. One important thing is that if you are looking to truly prove the value of AI, linking value to a quantifiable metric and, ideally, a financial one is very advantageous. Showing that AI has saved time or money sounds simple enough but this is often far more complex to measure than statistical metrics alone. In our experience, by defining value and then aligning on how you will quantify and measure the value, you set the foundation to build a strong AI business case. (For more tips on constructing a business case for new technology, check out our other blog.)

Here is a quick recap:

1. Define what you expect the value to be

2. Identify quantifiable metrics

3. Agree on how you will measure the metrics

How can I test if AI adds value?

The unhelpful answer to this question is that it depends. It comes down to what specific area you are looking to implement AI in, how you have defined value, and what you have defined as your quantifiable metrics.

In this article, to give an example, we will focus on the value of reducing cycle times in drug discovery. If achieved, fewer assays need to be run, so the quantifiable metric is reduced spending, and of course, we also have time savings to measure.

To oversimplify a typical drug development project, you start with a selection of chemical compounds that could potentially become new medicines. These could come from an existing pool or be designed based on existing information. Typically, these would be synthesised, tested, and analysed, which then creates an active feedback loop to learn from the test results to design more compounds to improve on their predecessors. The cycle is repeated as many times as necessary, getting to a point where compounds tick enough boxes to progress them confidently into development.

Testing how much value, in this case, time and cost saving, AI could deliver can be done in a wide variety of ways, and it will depend on the specific implementation of AI you are looking at. You could, in principle, do this either with a retrospective or prospective comparison. The problem with a prospective approach is that you don’t have a baseline against which to compare. In theory, you could run several projects twice, from the same starting points, with- and without the support of AI. That would give you a statistically significant sample from which to calculate the cost savings from AI. But, in practice, no one will ever do this! In our experience a retrospective analysis can often be most effective approach.

By applying AI to data from a project that has already been completed or is nearing completion, one can simulate how the project might have unfolded with AI’s involvement. While there are some limitations because the AI can only explore the compounds and data that were actually generated during the project, if you can isolate the original data set, you can relatively easily compare the number of compounds synthesised and the number of experimental measurements taken in the original project against an AI-assisted simulation of the same project.

Done correctly, this approach provides quantitative data that support the value proposition of AI and should give you a clear picture of whether AI could have helped speed up the process and reduce the need for extensive testing while still achieving high-quality results.

Recap:

1. Value: Reduce cycle times in drug discovery

2. Quantifiable metrics: time taken & number of experiments run

3. Retrospective analysis: repeat a completed project and compare metrics to the original

How long should it take?

When evaluating AI’s value in drug discovery, one crucial question is: How long does it take to demonstrate its impact? The answer is not as ambiguous as it may seem. It’s essential to establish a clear, time-bound evaluation period before starting any AI-related assessment.

Based on our experience, if you can’t establish a return on investment (ROI) business case within three to six months, we’d be cautious about moving forward. A three-month window is typically enough time to make tangible progress in your AI evaluation, allowing you to compare AI-driven simulations against real-world project timelines. It may stretch to a bit longer on occasion but if AI truly adds value, you should be able to see a noticeable reduction in time and effort spent on key processes within this period.

Setting a time-bound goal ensures that you can test AI’s impact in a controlled, practical setting, where you can measure improvements in speed, accuracy, and overall outcomes. This timeframe allows you to evaluate whether the technology meets your expectations and supports a compelling business case for its continued integration.

Conclusion

Proving the value of AI in drug discovery doesn’t need to be an overwhelming process. By taking a structured approach and aligning with what you aim to achieve, it becomes easier to demonstrate AI’s potential benefits. Start by clearly defining the value you’re seeking, and then establish a realistic testing framework. With these in place, testing AI’s value is not as taxing as it might first appear. By the end of the evaluation period, you should have a clear sense of whether AI can help improve your workflows, speed up development, and ultimately provide a strong return on investment.

Ready to transform to AI-guided drug discovery?

Simply complete the form and our team will be happy to chat about how Cerella could improve your discovery workflows.

About the author

James Halle

James leads Optibrium’s commercial team, overseeing global sales, customer success, and marketing while shaping the company’s product commercialization strategy. He brings extensive experience from senior roles at Cytel, IQVIA, IBM, and LexisNexis, combining technical and business expertise with a BSc in Computer Science from the University of Cardiff.

Linkedin

James Halle, Chief Commercial Officer (CCO)

More drug discovery resources