From Black Boxes to Actionable Insights: A Perspective on Explainable Artificial Intelligence for Scientific Discovery

Journal of Chemical Information and Modeling, Journal Year: 2023, Volume and Issue: 63(24), P. 7617 - 7627

Published: Dec. 11, 2023

The application of Explainable Artificial Intelligence (XAI) in the field chemistry has garnered growing interest for its potential to justify prediction black-box machine learning models and provide actionable insights. We first survey a range XAI techniques adapted chemical applications categorize them based on technical details each methodology. then present few case studies illustrate practical utility XAI, such as identifying carcinogenic molecules guiding molecular optimizations, order chemists with concrete examples ways take full advantage XAI-augmented chemistry. Despite initial success chemistry, we still face challenges developing more reliable explanations, assuring robustness against adversarial actions, customizing explanation different needs diverse scientific community. Finally, discuss emerging role large language like GPT generating natural explanations discusses specific associated them. advocate that addressing aforementioned actively embracing new may contribute establishing an indispensable technique this digital era.

Language: Английский

---

Interpretable Attribution Assignment for Octanol–Water Partition Coefficient

The Journal of Physical Chemistry B, Journal Year: 2023, Volume and Issue: 127(31), P. 7004 - 7010

Published: July 27, 2023

With the increasing development of machine learning models, their credibility has become an important issue. In chemistry, attribution assignment is gaining relevance when it comes to designing molecules and debugging models. However, attention only been paid which atoms are in prediction not whether reasonable. this study, we developed a graph neural network model, highly interpretable model modified integrated gradients method. The our approach was confirmed by predicting octanol-water partition coefficient (logP) evaluating three metrics (accuracy, consistency, stability) assignment.

Language: Английский

---

Linear Graphlet Models for Accurate and Interpretable Cheminformatics

Published: Feb. 26, 2024

Advances in machine learning have given rise to a plurality of data-driven methods for estimating chemical properties from molecular structure. For many decades, the cheminformatics field has relied heavily on structural fingerprinting, while recent years much focus shifted leveraging highly parameterized deep neural networks which usually maximize accuracy. Beyond accuracy, techniques need intuitive and useful explanations predictions models uncertainty quantification so that practitioner might know when model is appropriate apply new data. Here we show linear built unfolded molecular-graphlet-based fingerprints attain accuracy competitive with state art retaining an explainability advantage over black-box approaches. We how produce precise by exploiting relationships between graphlets these are consistent intuition, experimental measurements, theoretical calculations. Finally use presence unseen fragments molecules adjust quantify uncertainty.

Language: Английский

---

Enhancing predictive modeling of photovoltaic materials’ solar power conversion efficiency using explainable AI

Energy Reports, Journal Year: 2024, Volume and Issue: 11, P. 3824 - 3835

Published: March 26, 2024

We present a study on Explainable AI-based prediction of power conversion efficiency (PCE) organic solar cells, conducted dataset 566 small-molecule cell materials samples with varying donor and acceptor species combinations. This research uncovers an interesting phenomenon, the first its kind to be reported, PCE quantization, where values increase in steps feature values. Our findings have significant implications for development efficient as they provide better understanding factors that influence PCE, highlight value ranges which more would achieved. demonstrates XAI techniques uncovering hidden patterns scientific datasets highlights importance interdisciplinary field science.

Language: Английский

---

Explaining compound activity predictions with a substructure-aware loss for graph neural networks

Journal of Cheminformatics, Journal Year: 2023, Volume and Issue: 15(1)

Published: July 25, 2023

Abstract Explainable machine learning is increasingly used in drug discovery to help rationalize compound property predictions. Feature attribution techniques are popular choices identify which molecular substructures responsible for a predicted change. However, established feature methods have so far displayed low performance deep algorithms such as graph neural networks (GNNs), especially when compared with simpler modeling alternatives random forests coupled atom masking. To mitigate this problem, modification of the regression objective GNNs proposed specifically account common core structures between pairs molecules. The presented approach shows higher accuracy on recently-proposed explainability benchmark. This methodology has potential assist model pipelines, particularly lead optimization efforts where specific chemical series investigated.

Language: Английский

---