Drug Discovery AI & Automation |

Predicting Failure Before it Happens: Leveraging AI/ML in Drug Discovery

Farhan Khodaee, Chief Executive Officer

| Technology Showcase

28 November, 2025

Watch time: 5 Minutes

Farhan Khodaee delivered a presentation outlining the challenges faced in drug development, particularly the high rate of clinical trial failures. Khodaee began by acknowledging that, despite extensive optimisation, a significant proportion of drugs fail once they reach the clinic. He highlighted that at Absentia, the team is developing predictive models to identify potential failures before drugs even reach the clinical stage, aiming to save resources and prevent costly setbacks. Khodaee noted that approximately 90% of drugs fail in clinical trials, with phase one failures often attributed to toxicity or side effects, and phase two failures resulting from a lack of efficacy. He emphasised that over 40% of these failures are due to side effects or toxicities that were not predicted during preclinical testing. Traditional preclinical assessments, even after asset optimisation, frequently fail to provide insights relevant to clinical outcomes. Absentia’s approach involves leveraging comprehensive clinical data to inform early predictions of toxicity and side effects. Khodaee illustrated this with the example of tenidap, a drug developed by Pfizer for anti-inflammatory diseases, which failed in clinical trials due to kidney and liver toxicities, despite promising preclinical results. He explained that Absentia’s model analyses the molecular space and screens for properties such as whole-organism toxicity, which are often overlooked in conventional screening processes. The model integrates ADME properties to predict how a drug will impact human biology and generates detailed toxicology reports. It can specify the type of toxic effect expected, provide quantitative measures such as IC50 for specific organs, and simulate drug interactions at the tissue and cellular level. These insights are presented in IND-enabling reports, offering explanations for predicted toxicities and guidance on potential drug modifications to reduce adverse effects while maintaining therapeutic efficacy. Khodaee stated that Absentia’s system delivers rapid, cost-effective toxicity assessments, typically within 48 hours, relying solely on computational methods. The reports include clear reasoning for toxicity predictions, enabling researchers to anticipate and address issues before clinical investment. He concluded by inviting collaboration and further discussion, noting that Absentia’s work is a spin-off from doctoral research at MIT and continues to advance predictive toxicology in drug development.

0:00
Yes.

0:00
So I know that we're in a discovery conference, but a lot of the drugs that we actually create and we spend a lot of time optimising, they fail in the clinic.

0:12
So what we are building at Absentia is trying to predict these failures before even going to clinical stage and even before touching a pipette.

0:24
90% of drugs actually fail in the clinical trial.

0:27
This happened either in phase one because of toxicity or side effects or is phase two because of lack of efficacy.

0:34
Based on the data that we gather from all the clinical data that is happening, more than 40% of these failures are due to them some side effect or toxicity that we couldn't really predict in the preclinical stage.

0:48
Most of the testing that we do in preclinical, even after optimising our asset, they're not really telling us anything about the clinical stage.

0:56
That's what we are building absentia to inform these failures early on.

1:02
Besides that we are spending millions of dollars in the later stages in discovery, things are really cheap.

1:08
But when you go in preclinical and when you go to clinical, you're investing a lot of money on these drug assets.

1:14
So it's really, really important to predict them really early on before going and spending all these millions of dollars and making companies bankrupt.

1:27
So let me walk you through an example that we ran through our model.

1:30
I'm not going to tell you anything about like how we build this, but I'm going to tell you how we are outputting the drugs.

1:35
This is a case tenidap that was developed by Pfizer around 20 years ago.

1:40
It was really a, it was going to be a blockbuster, very potent drug for anti-inflammatory diseases, especially RA.

1:51
It went to clinical trial, and it failed significantly because of kidney and liver toxicities.

1:57
And this is despite the fact that all the preclinical studies like rat studies show that there's no toxic effect.

2:06
Have we when we just looking at tenidap across like all the molecular space is in a very special part of the molecular space.

2:13
And you know, when we are screening these molecules, we usually look at like certain properties that are important for us.

2:20
But toxicity especially like whole Organism toxicity is not something that we really screen for.

2:25
And that's what we do at absentia.

2:28
We look at all the ADME properties, and these are very important to understand how it drugged impact an organism like human, and we use that information to inform a toxicology report.

2:41
This is our model can say specifically what kind of toxic effect will be seen if these drugs interact with human body.

2:50
We give a specific, you know, numbers even like IC50 when it interacts with kidney, for example, or specific kidney cells based on all the previously done experiments and assay that have been conducted anywhere in the Internet.

3:08
And we also simulate what happens in every single tissue and every single cell type when this drug starts interacting with different transcription factors or pathways that are involved with the mechanism of action.

3:22
We report these things as a form of, you know, as you get like an IND enabling very detailed descriptions of how this drug is interacting with different cell types.

3:30
What is the what are the channels that's interacting and a lot of different things, different things.

3:35
And we provide visuals to guide how we can actually modify this drug to make it better.

3:39
In the case of the tenidap, which showed that we could have predicted those things.

3:43
And we even provided detailed modifications to the drug that can make it less toxic and still maintain a lot of therapeutic effect.

3:51
And we have rescuing a lot of the drugs actually failed in the clinic with our algorithm.

3:57
What we have built so far is looking at like a whole Organism sort of toxicity effect.

4:04
We can do it much faster.

4:06
We usually give our reports in less than 48 hours.

4:09
It's much cheaper because we are only restricted to computation.

4:13
And we usually provide our reports in the form of explanations that when we are saying something is toxic for liver, why we say that and what are the reasoning behind that.

4:24
So in this case, for example, we saw hepatotoxicity.

4:28
So we'll expecting like this will fail in Phase 1.

4:31
And that's exactly what happened with that with that drug.

4:35
If you're interested to learn more and work with us, we have a booth upstairs.

4:39
This is a spin off from my PhD research at MIT and my Co-founder’s.

4:43
We'll definitely be happy to talk with you and just, you know, learn about your programme and if we can help at any point.

Predicting Failure Before it Happens: Leveraging AI/ML in Drug Discovery

Absentia

More from this sponsor

What to watch next

You're just a click away