Explaining Deep Neural Network using Layer-wise Relevance Propagation and Integrated Gradients

Cik, Ivan; Rasamoelina, Andrindrasana David; Mach, Marián; Sinčák, Peter

doi:10.1109/sami50585.2021.9378686

Cited by 7 publications

(4 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given the availability of the actual knot core locations within knotted sequences, we conducted an evaluation to measure the model's capability to not only correctly classify an input sequence, but also to pinpoint the knot core's location. However, conventional interpretability techniques, like Layer Integrated Gradients (Cik et al, 2021 ), have limited applicability on biological data since they can only relate to individual input points (in our case amino acids), whereas protein folding requires the cooperation of groups of them simultaneously. Therefore, we proposed a patching technique: we monitored how the model's prediction changed after replacing a continuous segment of amino acids in the original sequence with X characters, with respect to the model's prediction of the original unpatched sequence.…”

Section: Models and Methodsmentioning

confidence: 99%

Knot or not? Identifying unknotted proteins in knotted families with sequence‐based Machine Learning model

Sikora,

Klimentova,

Uchal

et al. 2024

Protein Science

View full text Add to dashboard Cite

Knotted proteins, although scarce, are crucial structural components of certain protein families, and their roles continue to be a topic of intense research. Capitalizing on the vast collection of protein structure predictions offered by AlphaFold (AF), this study computationally examines the entire UniProt database to create a robust dataset of knotted and unknotted proteins. Utilizing this dataset, we develop a machine learning (ML) model capable of accurately predicting the presence of knots in protein structures solely from their amino acid sequences. We tested the model's capabilities on 100 proteins whose structures had not yet been predicted by AF and found agreement with our local prediction in 92% cases. From the point of view of structural biology, we found that all potentially knotted proteins predicted by AF can be classified only into 17 families. This allows us to discover the presence of unknotted proteins in families with a highly conserved knot. We found only three new protein families: UCH, DUF4253, and DUF2254, that contain both knotted and unknotted proteins, and demonstrate that deletions within the knot core could potentially account for the observed unknotted (trivial) topology. Finally, we have shown that in the majority of knotted families (11 out of 15), the knotted topology is strictly conserved in functional proteins with very low sequence similarity. We have conclusively demonstrated that proteins AF predicts as unknotted are structurally accurate in their unknotted configurations. However, these proteins often represent nonfunctional fragments, lacking significant portions of the knot core (amino acid sequence).

show abstract

Section: Models and Methodsmentioning

confidence: 99%

Knot or not? Identifying unknotted proteins in knotted families with sequence‐based Machine Learning model

Sikora,

Klimentova,

Uchal

et al. 2024

Protein Science

View full text Add to dashboard Cite

show abstract

“…For model-agnostic FI, SHAP 1 and LIME 2 are stateof-the-art open-source FI frameworks [21], [22]. Other FI frameworks such as integrated gradients and layer-wise relevance propagation are specific to neural networks and mainly used for image datasets [23].…”

Section: B Xai Frameworkmentioning

confidence: 99%

Optimal Neighborhood Contexts in Explainable AI: An Explanandum-Based Evaluation

Pawar,

O'Shea,

O'Reilly

et al. 2024

IEEE Open J. Comput. Soc.

View full text Add to dashboard Cite

Over the years, several frameworks have been proposed in the domain of Explainable AI (XAI), however their practical applicability and utility need to be clarified. The neighbourhood contexts are shown to significantly impact the explanations generated by XAI frameworks, thus directly affecting their utility in addressing specific questions, or "explananda". This work introduces a methodology that use a comprehensive range of neighbourhood contexts to evaluate and enhance the utility of specific XAI techniques, particularly Feature Importance and CounterFactuals. In this evaluation, two explananda are targeted. The first one examines whether features' collection should be halted as per the AI model based on the sufficiency of the current set of information. Here, the information refers to the features present in the data used to train the AI-based system. The second one explores what is the most effective information (features) that should be collected next to ensure that the AI outputs the same classification as it would have generated with all the information present. These questions serve as a platform to demonstrate our methodology's ability to assess the impact of customised neighbourhood contexts on the utility of XAI.

show abstract

“…Given the availability of the actual knot core locations within knotted sequences, we conducted an evaluation to measure the model's capability to not only correctly classify an input sequence, but also to pinpoint the knot core's location. However, conventional interpretability techniques, like Layer Integrated Gradients, 43 have limited applicability on biological data since they can only relate to individual input points (in our case amino acids), whereas protein folding requires the cooperation of groups of them simultaneously. Therefore, we proposed a patching technique: we monitored how the model's prediction changed after replacing a continuous segment of amino acids in the original sequence with X characters, with respect to the model's prediction of the original unpatched sequence.…”

Section: Interpretation With Patching Techniquementioning

confidence: 99%

Knot or Not? Sequence-Based Identification of Knotted Proteins With Machine Learning

Šrámková,

Sikora,

Uchal

et al. 2023

Preprint

View full text Add to dashboard Cite

Knotted proteins, although scarce, are crucial structural components of certain protein families, and their roles remain a topic of intense research. Capitalizing on the vast collection of protein structure predictions offered by AlphaFold, this study computationally examines the entire UniProt database to create a robust dataset of knotted and unknotted proteins. Utilizing this dataset, we develop a machine learning model capable of accurately predicting the presence of knots in protein structures solely from their amino acid sequences, with our best-performing model demonstrating a 98.5 % overall accuracy. Unveiling the sequence factors that contribute to knot formation, we discover that proteins predicted to be unknotted from known knotted families are typically non-functional fragments missing a significant portion of the knot core. The study further explores the significance of the substrate binding site in knot formation, particularly within the SPOUT protein family. Our findings spotlight the potential of machine learning in enhancing our understanding of protein topology and propose further investigation into the role of knotted structures across other protein families.

show abstract

Explaining Deep Neural Network using Layer-wise Relevance Propagation and Integrated Gradients

Cited by 7 publications

References 24 publications

Knot or not? Identifying unknotted proteins in knotted families with sequence‐based Machine Learning model

Knot or not? Identifying unknotted proteins in knotted families with sequence‐based Machine Learning model

Optimal Neighborhood Contexts in Explainable AI: An Explanandum-Based Evaluation

Knot or Not? Sequence-Based Identification of Knotted Proteins With Machine Learning

Contact Info

Product

Resources

About