The ONTOX hackathon: accelerating non-animal toxicology

In this blog, ATLA‘s Technical Editor Rita Seabra breaks down a recent paper about the first ONTOX Hackathon. The entire paper has been published in ATLA and is freely available. More information is also available on the ONTOX website

Who is ONTOX?

ONTOX is an EU-funded project developing a strategy for innovative, non-animal methods (NAMs) to predict how repeated exposure to chemicals affects the human body. By using AI-based computational systems fed by human physiological data and in vitro testing, the project focuses specifically on protecting the liver, kidneys, and developing brain.

The Problem: A growing chemical landscape

Currently, over 20,000 chemicals are registered under the EU’s REACH regulation. As this number increases, we’re inevitably exposed to complex mixtures that require quicker, more reliable safety assessments.

Traditional animal testing is often insufficient for this scale of data. As a result, there is a shift towards computer modelling and AI to provide more human-relevant risk assessments that both replace animals and offer more human-relevant results.

Why a hackathon?

Implementing these new methods comes with significant barriers. To address these “wicked problems”—complex issues with deep-seated societal and technical roots—the ONTOX project gathered a group of experts from regulatory authorities, industry, and animal welfare organisations.

The team decided that a “hackathon” was the most effective way to break through these hurdles. Usually found in the IT sector, a hackathon is an intensive event where participants work to deliver a functional prototype or solution within a very short timeframe (typically 48 hours).

The goal of the hackathon:

The central idea of the ONTOX hackathon was to explore the use of AI as an “accelerator” or “co-pilot” for non-animal testing, rather than a standalone solution. The 48 participants were split into eight groups to act as a “reality check” for the project’s strategy. They were tasked with solving four specific issues regarding the practical implementation of AI in toxicology, ensuring these new methods can move from theory into real-world use.

During the hackathon, they looked at four key challenges.

Challenge one: How to drive the use of AI in chemical risk assessment?

AI is expected to deliver substantial economic and societal benefits across many areas, including in chemical analysis and risk assessments. If AI is able to rapidly and efficiently process large volumes of data relevant for toxicological assessment with a high degree of confidence, it could potentially be a game-changer.

However, the algorithms underlying AI-based models for chemical analysis must be transparent and aligned with national laws and regulations. Importantly, mechanisms that enable data sharing while safeguarding intellectual property rights must be in place.

Participants were expected to address the drivers (e.g. explainability, trust, transparency, etc.) and the barriers (e.g. proprietary AI, mistrust, uncertainties, etc.) from a European and global perspective/market.

Challenge two: To predict or protect?

Toxicological testing has been hazard-oriented, as it predicts whether a substance is toxic or non-toxic after exposure. This approach doesn’t consider the possibility that substances might be used for different purposes. Thus, as the type of exposure varies, so does the probability of a negative effect. For example, if no exposure to a given chemical is expected, there is no significant risk and it would be possible to minimise the number of toxicology testing for that chemical. Thus, a risk assessment approach would consider several variables, e.g. likely exposure concentration, duration of exposure, known toxicity of the substance.

Participants were expected to consider whether it makes sense to change from the current hazard-oriented ‘prediction’ approach to a ‘protection’ approach which would then consider exposure in specific situations.

Challenge three: How can we secure human health and environmental protection at the same time?

The EU has a roadmap to become a climate-neutral region, the European Green Deal. This strategy aims to protect the environment while also ensuring access to food, safe and sustainable transport, energy, and industry.

A key question is whether protecting public health and safeguarding the environment are competing priorities, or whether they are aligned. The European Green Deal approaches human health and environmental protection in an integrated, holistic way, but there are challenges. For example, can healthy food be produced in a way that is economically viable without causing environmental harm?

Participants were asked to consider how to avoid situations in which efforts to protect human health and the environment come into conflict. They had to examine both driving factors like safer and more sustainable design approaches and potential obstacles like intellectual property constraints or limited expertise.

In addition, they were encouraged to identify the key stakeholders who should be involved in addressing these challenges.

Challenge four: How can we facilitate the transition from animal tests to full implementation of human relevant methods?

The EU citizens’ initiative ‘Save cruelty-free cosmetics — commit to a Europe without animal testing’ has pushed the EU Commission to speed up the implementation of non-animal methods for chemical substances. However, scepticism related to the uncertainty of tools like (AI-based tools or in vitro tests remains, even though traditional animal testing has similar uncertainties.

Today’s animal testing capacity is not enough to address society’s needs, but companies often have to carry out animal studies to comply with current regulations. Unfortunately, alternative test methods are fully accepted in only a few areas. In addition, regulators have usually demanded that non-animal methods be validated by comparing them to animal studies before approving them. The basic assumption is that industry will only adopt non-animal methods widely once regulators fully accept these methods.

Participants were asked to discuss how to drive the transition from animal-based chemical testing to the full implementation of human-relevant methods. They were expected to consider the drivers like addressing the incentives to chemical risk assessment and the barriers like current guidelines hampering the transition.

They were also asked to consider the following question: If we need to change the current approach to risk assessment, how will it be possible to enforce the equal validation of traditional animal in vivo tests and NAMs, with regard to the uncertainties inherent in both?

A couple of the proposals presented at the hackathon were going to be followed up, as they had potential to support the implementation and/or sustainability of an AI-based approach to risk assessment. However, the hackathon was not only about the development and implementation of complicated technical solutions to a problem, but also about how society feels about the risks posed by chemical substances. What level of uncertainty are we willing to accept? Can we trust something we do not fully understand? We also want to protect nature, animals and our own health to the maximum extent — are these conflicting goals?