Response to "Addressing biases and limitations in feature attribution for circRNA modification profiling".

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Circular RNAs (circRNAs) play crucial roles in gene regulation.
• Accurate profiling of circRNA modifications is essential for understanding their functions.
• Feature attribution methods are increasingly used to interpret complex biological data.

Purpose of the Study:

• To clarify the methodologies used in a previous study on circRNA modification profiling.
• To address concerns regarding biases and limitations in feature attribution techniques.
• To differentiate the roles of independent models in prediction and interpretation.

Main Methods:

• Utilized two independent XGBoost models: one for predicting RNA modification events from nanopore signal features, and another for feature attribution using genome-derived sequence features via the m6AlogisticModel framework.
• Employed Shapley Additive Explanations (SHAP) as an exploratory tool for model interpretability.
• Responded to specific comments regarding methodological choices and their implications.

Main Results:

• Confirmed the distinct applications of the two XGBoost models in the analysis.
• Clarified that SHAP was used for exploratory purposes, not to establish causal biological mechanisms.
• Acknowledged the value of constructive methodological suggestions for future research.

Conclusions:

• The study reaffirms the appropriate use of XGBoost models for circRNA modification analysis.
• Highlights the importance of understanding the limitations of interpretability tools like SHAP.
• Suggests that integrating complementary analytical strategies can improve the robustness of computational studies in circRNA modification profiling.