SciPoll 396: SOT/EUROTOX Debate: Is There a Role for Artificial Intelligence (AI) and Machine Learning (ML) in Risk Decisions?
On a scale of 1 to 5 (1=Strongly do not support; 3=equivocal, 5=strongly support), how strongly do you support that AI and ML can be applied in toxicology as they are being applied in precision medicine and other fields?
Results
(132 Answers)
Answer Explanations 55
I believe that these tools can be used in conjunction with human intelligence for risk assessment. But, many a times we find false negative with these tools, still experiments (in-vitro, in-vivo) are required. These tools can reduce the number of experiments by utilizing techniques like read-across, QSAR, Toxicokinetic modelling etc.
Risk prediction/identification could be the most important.
I think they could be used a screen, but need review
AI and ML are tools that can be very helpful IF properly used. Toxicology is no different than other scientific areas. Data are collected and analyzed.
AI and MA significantly contributed to the personalized medicine, however, human is complex, and the factors involved in toxicology, particularly in human are multifactorial
Recent research articles have strongly demonstrated the predictive power of machine learning approaches for various aspects of toxicological hazard based on chemical structure and other existing bioactivity data.
Sometimes the probability calculations in multiple risk variables that may of may not interact, are very large and AI , also possibly ML, could be used to estimate probability.
Expedient and accurate results may be obtained
AI provides complementary data for decision and is equally accurate to or even better than animal mehods
AI already plays a major role in risk evaluation: In uncertainty analysis, benchmark dose modeling, combined and weighted risk of exposure to mixtures, exposure assessment, scenario assessments, toxicokinetics modeling, QSAR, read-across
I feel that AI amd ML could be game changers in the future but their output totally depends upon the quality and depth of input data.
ML may repr3sent the future of precision medicine, surely helping in decision making progress and to ensure fster and safer care for patients.
However, I believe that it s probably too early at the moment for giving AI too much responsibilities
ML and AI offer unprecedented levels of risk validation options, particularly in combining multiple algorithms like has never happened before. This is bound to be superior than single testing methods as were used in the past (still done) and at least as good as wet lab testing but without the need for invasive analysis.
There many areas in toxicology where AI and ML can be conveniently be useful
If datasets are available, as they may be for various compounds used in precision medicine, in addition to the genotypes of individuals (pharmacogenetics, having a gene that may be an appropriate target for therapy), then AI and ML may be useful in predicting efficacy and safety in humans.
Assessment of toxicology involve a systems level involvement; including the enzymatic reactions within host and microbiome. Thus, if these components are not incorporated in the test, the reliability of output is questionable.
ML and AI are simply new statistical tools that can be helpful or harmful depending on the assumptions and interpretation. No different from a t-test!
Conceptually they can be applied equally. However, given the uncertainties about mechanisms of toxicity even though the promise is there it will take time and experience to be widely applied with confidence
Still alot about tox we dont know
Certainly we can teach computers to calculate LD50s and EC50s as well as DR modeling
I think these techniques can add much information but need to be used carefully.
Prediction by AI should be properly verified experimentally in a preliminary stage. The models should be properly built and verified.
Precision medicine uses information about a person's own genes, proteins or images to prevent, diagnose, or treat disease in comparison to large population data bases of healthy and diseased/injured persons. Toxicology presently does not have that data base or anything remotely close for most of the in vitro/stem cell/organ-on-a-chip/etc. technologies in use. The positive controls for these experiments are rather limited and do not inform about many biological/biochemical/molecular pathways, let alone about Phase 1, 2 and 3 metabolism and organ interactions or the blood-brain-barrier. On the other hand, this is a lofty goal through which continued efforts may bear useful fruit
Reductions of experiments to be done
In an individual research, despite enormous effort, we can not accommodate lots of variables, mega sample size, and complicated modeling/ analysis. Hence Artificial Intelligence (AI) and Machine Learning (ML) can be useful to accommodate lots of variables, mega sample sizes, and complicated modeling/ analysis.
This is still an emerging field, we will have a better understanding in the next few years. I think it will be as we use QSAR applications now. It will definitely reduce animal testing significantly, will it eliminate it altogether, I think not; we will still need a few final studies in animals for a better risk assessment and decisions.
I believe it is helpful to use all tools that are available. In particular, I feel that AI and ML have applications in simulation and statistical analysis models.
The introduction of artificial intelligence and machine learning will speed up the process of toxicological characterization of chemicals, particularly in the first stage as screening tools. However, the advancement of knowledge in this field would imply a continuous review of the algorithms. No final decision shall be taken based solely on the AI process.
First of all, the concept of "precision medicine" is ill-defined and seems to be more of a fad to access research funding. Furthermore, one often quoted concept, i.e. "Precision medicine provides a model for the next generation of lifestyle modification." seems not really relevant for toxicology and risk assessment.
Finally, all the points addressed in "Ethical and legal challenges of artificial intelligence-driven healthcare" (Gerke et al., Artificial Intelligence in Healthcare. 2020 : 295–336) would apply to toxicology as well.
In conclusion, I do not see an immediate future for AI or ML in toxicology.
Only 4 due to potential risks associated with ADS.
See: EPRS | European Parliamentary Research Service Scientific Foresight Unit (STOA) PE 624.261 – March 2019 EN
Understanding algorithmic decision-making: Opportunities and challenges
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwiUv8-bwtL2AhWOUMAKHQddAYgQFnoECAoQAQ&url=https%3A%2F%2Fwww.europarl.europa.eu%2Fthinktank%2Fen%2Fdocument.html%3Freference%3DEPRS_STU(2019)624261&usg=AOvVaw10oFIqtpDMg8UdDyjol1B-
AI and ML can uncover mechanistic pathways in risk
Despite tremendous investigation on the complex biology of human being and the fact that immune system is capable to defend against much toxic chemicals, the knowledge gaps in our understanding make complicate the accurate risk assessment formulation if AI and ML were applied to toxicological studies.
Seems like it's not a matter of "can be". Rather, they are tools that must be considered so that we do not continue to waste the value, and continue to delay discoveries (about what is toxic as well as what is not), in the huge preclinical and mechanistic data sets we can now generate.
Animal testing is under a great deal of pressure for the last 2 decades or more. Big data and AI have come a long way in leveraging existing animal data for tox assessment.
Of course this application should be attempted, but it's success is not assured and will required (extreme) validation.
AI and ML can identify complex patterns and relations in compllex systems and have been demostrated to be very reliable on both prediction and diagnosis.
Sensori ed AI sono una reciproca integrazione (tipo hardware/software)preziosa per monitorare tossicità
I think there is a huge potential for AI and ML for (eco)toxicology.
Such approaches can broaden the insights that can be derived from high-throughout testing including single-cell technologies
AI/ML is nothing different than theories/models executed by a computer. Making those models/theories executable is non-trivial, but our perception of AI/ML should not mix that of AI/ML doing predictions and AI/ML as tool during research to help explore and understand patterns.
Distinct strengths in approach but careful evaluation of real-world model performance and application is needed
Need to accurately mirror the animal toxicological data and should not be used in isolation unless verification and validation (i.e., predictability) have occurred.
The risk profile from what I have read seems to be minimal compared to net benefits.
AI is a valuable tool especially for "Tier 1" or screening level assessment of toxicology and risk.
AI and ML have use in all fields of academics. Toxicology is one of them. Starting form dose drug determination to effects evaluation AI and ML can be useful.
There are lots of data, knowledge in very different formats which must be meaningful and valuable for toxicology predictions, however difficult for human review. It is exactly the space where AI and ML could and should play a role to identify patterns, signals hidden behind data etc. They could also provide valuable predictions on risk mitigation measures.
AI and ML may be used as to an aid support as part of a WoE but not to provide a definitive answer.
It will be a challenge humanity needs to overcome. With big data (GPS, -omics, and the necessity of learning in a multiple stressor (chemical and nonchemical exposures)… machine
Learning is unavoidable.
I have used numerous programs (e.g. Leadscope, In Vitro MultiFlow) and none have accurately predicted the actual biological effect.
AI and ML are just tools and they often help us to look at the data in different ways. We are not obligated to use the results, but why wouldn't we want to find patterns that our own eyes can't see?
Although I admit that AI and ML might be useful tools to go through huge piles of data, I have doubts about whether or not it is really desirable to have such tools. Even with our PC we often can see that they are living their own life upon which we don´ t have control any more. And, as Douglas Hofstadter pointed out, if AI reaches human intelligence, it will be as pone of errors as human brain is.
AI and ML can more quickly analyse a massive amount of data to find a signal than the human mind.
I don't think AI and ML are ready for toxicology for Risk assessment, or perhaps it should be framed that toxicology is not ready for AI/ML yet as it requires large amount of reproducible data. And due to the variabilities nature of animal/human studies and complexities of biological systems, until we have consistent reproducible methods (in vitro, NAMs, AOPs), AI/ML would not be practical in toxicology.
AI and ML processes will support toxicology analysis in the future.
As datasets get larger and more complicated, machine learning methods are strongly positioned to disentagle and identify relationships that would otherwise not have been identified. If nothing else, it serves as an important tool in hypothesis generation.
user-750950
05/22/2023 13:41I think there is a high potential forML and AI for (eco)toxicology.