Feature Importance in Predicting Clinical Outcome: Statistics vs. Explainable Artificial Intelligence
Amin, P.
Show abstract
At the time of diagnosis for cancer patients, a wide array of data can be gathered, ranging from clinical information to multiple layers of omics data. Determining which of these data are most informative is crucial, not only for advancing biological understanding but also for clinical and economic considerations. This process facilitates the selection of the most significant markers, enhancing patient stratification and informing treatment recommendations. In this paper, we start with 89 features extracted from multiomics and clinical data and aim to identify the most important ones in predicting response to neoadjuvant chemotherapy (NAC) using different explainable Artificial Intelligence (XAI) models and statistics. Our results show that XAI methods consistently recover important features that are missed by statistics and vice versa, hinting towards the need for complementary implementation of these methods. Furthermore, we find that a myriad of features, from mutations to immune infiltration, affect the response to NAC in breast tumors.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Development of an absolute assignment predictor for triple-negative breast cancer subtyping using machine learning approaches 95%
- BenchXAI: Comprehensive Benchmarking of Post-hoc Explainable AI Methods on Multi-Modal Biomedical Data 94%
- Deep Multimodal Graph-Based Network for Survival Prediction from Highly Multiplexed Images and Patient Variables 93%
Similar papers in this journal
- Stochastic LASSO for extremely high-dimensional genomic data 95%
- Accurate Prediction of Breast Cancer Survival through Coherent Voting Networks with Gene Expression Profiling 95%
- DeepInsight-3D for precision oncology: an improved anti-cancer drug response prediction from high-dimensional multi-omics data with convolutional neural networks 95%
Similar papers in this journal
- A variational autoencoder trained with priors from canonical pathways increases the interpretability of transcriptome data 94%
- Highly Accurate Cancer Phenotype Prediction with AKLIMATE, a Stacked Kernel Learner Integrating Multimodal Genomic Data and Pathway Knowledge 94%
- Methodological Challenges in Translational Drug Response Modeling in Cancer 94%
Similar papers in this journal
Similar papers in this journal
- Leveraging Permutation Testing to Assess Confidence in Positive-Unlabeled Learning Applied to High-Dimensional Biological Datasets 96%
- Probabilistic modeling methods for cell-free DNA methylation based cancer classification 94%
- GEOlimma: Differential Expression Analysis and Feature Selection Using Pre-Existing Microarray Data 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.