Garbage in, garbage out: mitigating risks and maximizing benefits of AI in research

Hanson, Brooks; Stall, Shelley; Cutcher-Gershenfeld, Joel; Vrouwenvelder, Kristina; Wirz, Christopher; Rao, Yuhan; Peng, Ge

doi:10.1038/d41586-023-03316-8

Cited by 15 publications

(4 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The accuracy of the values predicted by an ANN model is relatively high for many cases ( Figure 2 ) [ 69 ]. However, if the training experimental data contain considerable error, the trained ANN model becomes inaccurate, for which it is said garbage in , garbage out [ 77 ]. The interpretability of an ANN model is relatively low because an ANN model is very complex from the viewpoint of simple causality [ 78 ].…”

Section: Machine Learning (Ml)mentioning

confidence: 99%

Merits and Demerits of Machine Learning of Ferroelectric, Flexoelectric, and Electrolytic Properties of Ceramic Materials

Yasui

2024

Materials

View full text Add to dashboard Cite

In the present review, the merits and demerits of machine learning (ML) in materials science are discussed, compared with first principles calculations (PDE (partial differential equations) model) and physical or phenomenological ODE (ordinary differential equations) model calculations. ML is basically a fitting procedure of pre-existing (experimental) data as a function of various factors called descriptors. If excellent descriptors can be selected and the training data contain negligible error, the predictive power of a ML model is relatively high. However, it is currently very difficult for a ML model to predict experimental results beyond the parameter space of the training experimental data. For example, it is pointed out that all-dislocation-ceramics, which could be a new type of solid electrolyte filled with appropriate dislocations for high ionic conductivity without dendrite formation, could not be predicted by ML. The merits and demerits of first principles calculations and physical or phenomenological ODE model calculations are also discussed with some examples of the flexoelectric effect, dielectric constant, and ionic conductivity in solid electrolytes.

show abstract

Section: Machine Learning (Ml)mentioning

confidence: 99%

Merits and Demerits of Machine Learning of Ferroelectric, Flexoelectric, and Electrolytic Properties of Ceramic Materials

Yasui

2024

Materials

View full text Add to dashboard Cite

show abstract

“…In practice, however, this astranomical task is computationally impossible, which explains partly why this article puts forward just only a structural biophysics-based prototype (i.e., semaGIBAC), instead of a real GIBAC. Therefore, this article here again proposes an AI-based open strategy [95] to make it possible and conceivable to generate a vast amount of data to train a real GIBAC with reasonable accuracy and precision, as openness in data acquisition and generation, and AI algorithms is essential for promoting transparency, reproducibility, and collaboration within the community of drug discovery & design, and for facilitating continued improvement of the performance (accuracy, precision and efficiency) of GIBAC in precise drug discovery & design [35,36] in future. As mentioned above, this article puts forward a set of in silico steps of structural and biophysical data generation towards a paradigm shift in precise drug discovery and design [59,60].…”

Section: In Silico Generation Of Structural and Biophysical Data With...mentioning

confidence: 99%

A one-dimensional semaglutide-GLP-1R-based mini static GIBAC

2024

Preprint

View full text Add to dashboard Cite

Drug-target binding is an essential parameter in drug discovery and design to ensure drug efficacy and specificity. In 2022, the concept of a general intermolecular binding affinity calculator (GIBAC) was for the first time coined: $Kd = f(molecules, envPara)$. Technically, GIBAC represents a set of in silico approaches for structural and biophysical data generation towards a paradigm shift in precise drug discovery and design, providing a comprehensive framework for intermolecular binding affinity calculation with adequate accuracy, precision and efficiency. For the first time, this study reports a prototype of GIBAC (semaGIBAC), i.e., a one-dimensional semaglutide-GLP-1R-based mini static GIBAC, based on an experimental complex structure of semaglutide and GLP-1R. Semaglutide is a potent GLP-1 receptor agonist used to treat type 2 diabetes mellitus by regulating blood glucose levels and promoting weight loss. In 2021, a structural modification involving a Val27-Arg28 exchange was manually introduced to enhance semaglutide-GLP-1R binding affinity. This study employs a comprehensive structural and biophysical analysis aimed at thoroughly exploring the sequence space of semaglutide-GLP-1R to design analogues with improved binding affinity, leading to the identification of a promising semaglutide analogue, which binds to GLP-1R with an affinity that is more than two orders of magnitude (113.3 times) higher than native semaglutide. To sum up, this article puts forward a promising structural biophysical approach for developing GLP-1 receptor agonists with enhanced efficacy, and with a GIBAC prototype (semaGIBAC), this article argues again that the time is now ripe for the construction of a real GIBAC to be listed on the agenda of the drug discovery and design community.

show abstract

“…This has led to missing values in multi‐source data integration (garbage in, garbage out). The academic community has formed a consensus regarding the significant issues of similar data quality [29,30] . The issue of missing data serves as a common problem among multi‐source data, [31] as demonstrated in research fields such as medicine to reduce model performance [32,33] .…”

Section: Introductionmentioning

confidence: 99%

An ensemble learning strategy for multi‐source hydrogen embrittlement data by introducing missing information

Gong,

Lei,

Sun

et al. 2024

Materials Genome Engineering Advances

View full text Add to dashboard Cite

Accurately and quickly predicting hydrogen embrittlement performance is critical for the service of metal materials. However, due to multi‐source heterogeneity, existing hydrogen embrittlement data are missing, making it impractical to train reliable machine learning models. In this study, we proposed an ensemble learning training strategy for missing data based on the Adaboost algorithm. This method introduced a mask matrix with missing data and enabled each round of training to generate sub‐datasets, considering missing value information. The strategy first trained a subset of features based on the existing dataset and a selected method and continuously focused on the combination of features with the highest error for iterative training, where the mask matrix of the missing data was used as the input to fit the weights of each base learner using a neural network. Compared with directly modeling on highly sparse data, the predictive ability of this strategy was significantly improved by approximately 20%. In addition, in the testing of new samples, the predicted mean absolute error of the new model was successfully reduced from 0.2 to 0.09. This strategy offers good adaptability to the hydrogen embrittlement sensitivity of different sizes and can avoid interference from feature importance caused by filling data.

show abstract

Garbage in, garbage out: mitigating risks and maximizing benefits of AI in research

Cited by 15 publications

References 13 publications

Merits and Demerits of Machine Learning of Ferroelectric, Flexoelectric, and Electrolytic Properties of Ceramic Materials

Merits and Demerits of Machine Learning of Ferroelectric, Flexoelectric, and Electrolytic Properties of Ceramic Materials

A one-dimensional semaglutide-GLP-1R-based mini static GIBAC

An ensemble learning strategy for multi‐source hydrogen embrittlement data by introducing missing information

Contact Info

Product

Resources

About