Key features overviewΒΆ
The Signatures molecular descriptor [1-3] for computing descriptors from chemistry.
ECFPs of diameter 0-6 and physico-chemical descriptors computed using CDK.
Transductive Conformal Prediction (TCP) for binary and multi-label classification.
Inductive Conformal Prediction (ICP) for binary and multi-label classification, as well as regression.
Aggregated Inductive Conformal Prediction (ACP) according to [4] for binary and multi-label classification, as well as regression.
Cross-conformal Prediction (CCP) according to [8] for binary and multi-label classification, as well as regression.
Calculation of p-values using standard approach, smoothed and linear/splines interpolation, described in [11] and [12].
Cross Venn-ABERS Prediction (CVAP) according to [9] for probabilistic (binary) classification.
Significant Signature calculation and feature highlighting according to the Gradient method described in [5].
Parameter tuning using exhaustive grid search, using cross-validation or single test-train splitting, otherwise using default values according to [6].
LIBLINEAR and LIBSVM for training machine learning models used in CP/Venn-ABERS.