Advancements in Computational Toxicology: Integrating Big Data and AI

by

Increased efforts in the last few years towards computational toxicology, big data, and artificial intelligence have transformed the contemporary mechanism to evaluate toxic effects on human beings and the environmental implications of chemical substances. Acute toxicity testing and other common approaches to toxicity testing are gradually moving away from animal testing and drawing-based experimental tests and have moved a long way toward sophisticated computer-based algorithms that are capable of analyzing huge amounts of data in a far more accurate and precise manner than the traditional methods. This shift not only reduces animal testing and experimentation but also speeds up the identification of dangerous chemicals to construct safer products and help in forming better regulations. The adoption of big data and AI into this field is therefore an important step for computational toxicology to improve the understanding of chemical safety more ethically, efficiently, and in a much cheaper manner.

The Role of Big Data in Computational Toxicology

Toxicogenomics is one of the biggest fields of modern computational toxicology. Toxicity prediction and risk assessment of chemicals: modern opportunities due to the volume, variety, and velocity of data generated by high-throughput screening assays, omics technologies, and environmental monitoring systems. Nevertheless, maintaining and analyzing such a huge volume of information becomes possible only with the help of complex computation facilities and algorithms.

Such techniques as high-throughput screening assays, for instance, produce massive data sets that record how thousands of chemicals impact biological systems. These datasets are extremely useful in predicting any toxicants and better understanding how such compounds work. Though the most common nature of data is the inclusion of factors, which include concentrations, exposure time, and biological effects, this calls for the use of analytic tools of higher order. Big data solutions help to cope with such a situation because they allow identifying the dependencies and patterns that other methods would ‘overlook’.

Further, the practice of incorporating multiple platforms like genomics, transcriptomics, proteomics, and metabolomics offers a broader perception of how chemicals influence and are grasped by biological mechanisms. It is becoming increasingly clear that this integration of multi-omics and big data will uncover novel pathways of toxicity in addition to providing identification of biomarkers for early toxicity prediction. Due to the advances in big data, computational toxicology has been progressing to become more predictive and even more precise.

Yearwise Publication Trend on “computational toxicology”

Find publication trends on relevant topics

Artificial Intelligence in Toxicity Prediction

Computational toxicology is highly benefited from the developments of AI, especially from ML and DL approaches. These approaches are learned using training data and are supposed to work on high-dimensional data, and such a classifier is perfectly suitable for the classification of chemical compounds as toxic or non-toxic based on their molecular descriptors and biological activities.

Random forests, support vector machines, and neural networks have become popular in toxicity prediction. Some of these models can learn from raw data to find relationships between the structures of chemicals and their toxicological profiles. For example, ML algorithms can identify how probable it is that a single chemical substance will be toxic within certain parameters, e.g., carcinogenicity or endocrine modulation from the chemical’s structure. This capability of predicting with a high level of reliability without having to test the new compounds in experiments is a considerable advantage, especially when having to determine the toxicity of new or unknown chemicals.

Machine learning has advanced in computational toxicology through a sub-discipline known as deep learning. CNN and RNN deep learning models are particularly suited for processing and analyzing different data, such as images and sequences. Deep learning can be used in toxicity prediction to simulate the chemical-biological system at different hierarchy levels, from molecular interactions to cellular effects. As with many hierarchy learning techniques, deep learning models of chemical interactions will provide better interpretation and prediction of chemical toxicity.

Integrating Big Data and AI for Enhanced Toxicity Assessment

The commencement of big data and AI is considered a major progress in the field of computational toxicology. In general, using the endless sources of data and the abilities of AI to make predictions, scientists can create a more accurate and detailed model of chemical toxicity. This integration is highly relevant in the context of both HTS and multi-omics data analysis, as data complexity and volume become high and cannot be analyzed with standard tools.

It means that one of the major benefits of big data and AI integration is the possibility of building models that are both accurate and explanatory. Current in vitro and in vivo models involve a large number of statistical endpoints, which have a purely quantitative basis and do not give very appropriate qualitative pictures of toxicity. whereas other machine learning models can take mechanistic data like gene expression profiles and protein-protein interactions to give a clear picture of how the chemicals tested can physically harm. This mechanistic understanding is very helpful for the discovery of toxicity biomarkers and the design of specific countermeasures against the detrimental effects of chemicals.

Another important benefit that can be obtained from this integration is the sorting of chemicals, which might be considered for further study and testing according to their estimated toxicity. Since there are many thousands of chemicals currently being used in commerce and even more being synthesized, it would not be practical to screen each one in isolation by conventional laboratory methods. AI models can quickly predict drug candidates that are most toxic and so rule out large chemical libraries, thereby saving time for experts to work on more dangerous compounds. Such prioritization also speeds up the process of chemical safety assessment but, in the same manner, decreases the expenses and questionable moral foundations of animal testing.

Recent Publications on “computational toxicology”

Find publications on relevant topics

Case Studies: AI and Big Data in Action

Some of the very recent works have shown AI and big data can help bring a new level of improvement to computational toxicology. For instance, deep learning models have been used in the prognostic or prediction of hepatotoxicity of drugs according to their molecular features and pharmacological properties. Through such approaches, researchers have been able to develop models that, through training on large datasets of identified toxic and non-toxic chemical compounds, have been able to attain high levels of accuracy in diagnosing drug-induced liver injury (DILI), a common cause of drug pull-outs from the markets.

In another study, the developmental toxicity of chemicals was also predicted by using the artificial intelligence-driven models that were obtained from high-screening assay data and chemical database structure. It was possible to use these models to determine the features of molecules that are linked with developmental toxicity and to understand the processes of their action. The potential to estimate developmental toxicity without involving animals is crucial since chemical testing on pregnant animals is considered unethical and/or prohibited.

In addition, work has been done using AI to look at the dose-response relationships of quantified chemical exposures and associated toxicity, including health risks. AI integration of data from various sources, for example, environmental monitoring data, toxicogenomics data, and data from epidemiological studies, can help AI models discover new associations between chemical exposures and health risks. This could revolutionize environmental health research since it is likely to produce accurate and timely risk estimates of chemical exposure.

Challenges and Future Directions

However, some critical issues have not been fully solved in computational toxicology, promoted by big data and AI. This brings us to the first of some significant problems: the quality and consistency of the data that feeds AI. Information concerning toxicity is frequently ambiguous and cannot be classified as fully scientific because this information can be obtained through highly scientific toxicological experiments or wild guesses. This variability can pose challenges when developing AI models that are used for the prediction of the toxicity of new or unknown chemicals. To tackle this challenge, there is a need to keep up. with standard procedures for data curation and validation, plus the implantation of high-quality experimental data.

They are as follows: There is a major concern with explainability and interpretability, which is the ability to explain AI models. A significant concern with using AI-derived models of various diseases, pathogens, or disorders is that these models can paint a highly accurate picture of what might happen but are not very good at pointing out the reasons why things are likely to play out in a particular way. This lack of interpretability can be a problem for the implementation of AI in the regulation field since the decision-making process requires explanations. To counter this problem, scholars are coming up with new ways of explaining the results of an A.I. model, including feature importance analysis and result visualization.

As for future development, it could be expected that the enhancement and implementation of big data and AI into computational toxicology will bring more progress to existing and novel ideas. The advancement of new and complex models of AI that will be able to take and integrate multiple data forms and mimic complex biological systems will enhance mechanistic and robust predictive analysis of chemical toxicity.

In conclusion, big data and AI integration into computational toxicology is a giant leap in the advancement of the field, as it provides new and additional ways on how chemical safety assessment can be enhanced and how public health can be better guarded. These technologies will become fundamental as the future of toxicology depends on their progressive improvement as better ways of considering the effects of chemicals in the environment.

References

1. Davis AP, Wiegers TC, Johnson RJ, Sciaky D, Wiegers J, Mattingly CJ. Comparative Toxicogenomics Database (CTD): update 2023. Nucleic Acids Res. 2023 Jan 6;51(D1):D1257-D1262. doi: 10.1093/nar/gkac833. PMID: 36169237; PMCID: PMC9825590.

2. Ciallella HL, Russo DP, Sharma S, Li Y, Sloter E, Sweet L, Huang H, Zhu H. Predicting Prenatal Developmental Toxicity Based On the Combination of Chemical Structures and Biological Data. Environ Sci Technol. 2022 May 3;56(9):5984-5998. doi: 10.1021/acs.est.2c01040. Epub 2022 Apr 22. PMID: 35451820; PMCID: PMC9191745.

3. Ciallella HL, Russo DP, Aleksunes LM, Grimm FA, Zhu H. Revealing Adverse Outcome Pathways from Public High-Throughput Screening Data to Evaluate New Toxicants by a Knowledge-Based Deep Neural Network Approach. Environ Sci Technol. 2021 Aug 3;55(15):10875-10887. doi: 10.1021/acs.est.1c02656. Epub 2021 Jul 25. PMID: 34304572; PMCID: PMC8713073.

4. Xu Y, Dai Z, Chen F, Gao S, Pei J, Lai L. Deep Learning for Drug-Induced Liver Injury. J Chem Inf Model. 2015 Oct 26;55(10):2085-93. doi: 10.1021/acs.jcim.5b00238. Epub 2015 Oct 13. PMID: 26437739.

5. Zhu H. Big Data and Artificial Intelligence Modeling for Drug Discovery. Annu Rev Pharmacol Toxicol. 2020 Jan 6;60:573-589. doi: 10.1146/annurev-pharmtox-010919-023324. Epub 2019 Sep 13. PMID: 31518513; PMCID: PMC7010403.

6. Schwalbe N, Wahl B. Artificial intelligence and the future of global health. Lancet. 2020 May 16;395(10236):1579-1586. doi: 10.1016/S0140-6736(20)30226-9. PMID: 32416782; PMCID: PMC7255280.

7. Mayr, A., Klambauer, G., Unterthiner, T. and Hochreiter, S., 2016. DeepTox: toxicity prediction using deep learning. Frontiers in Environmental Science, 3, p.80.

8. Knapen D, Angrish MM, Fortin MC, Katsiadaki I, Leonard M, Margiotta-Casaluci L, Munn S, O’Brien JM, Pollesch N, Smith LC, Zhang X, Villeneuve DL. Adverse outcome pathway networks I: Development and applications. Environ Toxicol Chem. 2018 Jun;37(6):1723-1733. doi: 10.1002/etc.4125. Epub 2018 May 7. PMID: 29488651; PMCID: PMC6004608.

9. Ciallella HL, Zhu H. Advancing Computational Toxicology in the Big Data Era by Artificial Intelligence: Data-Driven and Mechanism-Driven Modeling for Chemical Toxicity. Chem Res Toxicol. 2019 Apr 15;32(4):536-547. doi: 10.1021/acs.chemrestox.8b00393. Epub 2019 Mar 25. PMID: 30907586; PMCID: PMC6688471.

10. Wu Y, Wang G. Machine Learning Based Toxicity Prediction: From Chemical Structural Description to Transcriptome Analysis. Int J Mol Sci. 2018 Aug 10;19(8):2358. doi: 10.3390/ijms19082358. PMID: 30103448; PMCID: PMC6121588.

Top Experts on “computational toxicology“