For the function _transformFeatures in model/mlLGPR.py starting from line 145 I think there might be a possible bug with how the pathway evidence features are extracted.
For example the idx for the classLabel "ACETOACETATE-DEG-PWY" is 670.
In line 148 featureNonItem[:, idx:idx + itemEvidenceFeaturesSize] this will extract the EC abundances from 670 to 702 (670 + 32).
Looking at S2 "Appendix. Features used for mlLGPR" describing the pathway evidence features it does not seem to be about EC abundances so I don't think it is extracting the pathway evidence features.
My guess would be that it should be something like:
ab_re_pp = self.nTotalComponents + self.nReacEvidenceFeatures + self.nPossibleClassFeatures
featureNonItem[ : , ab_re_pp + (idx * itemEvidenceFeaturesSize) : (idx * itemEvidenceFeaturesSize) + itemEvidenceFeaturesSize]
For the function _transformFeatures in model/mlLGPR.py starting from line 145 I think there might be a possible bug with how the pathway evidence features are extracted.
For example the idx for the classLabel "ACETOACETATE-DEG-PWY" is 670.
In line 148 featureNonItem[:, idx:idx + itemEvidenceFeaturesSize] this will extract the EC abundances from 670 to 702 (670 + 32).
Looking at S2 "Appendix. Features used for mlLGPR" describing the pathway evidence features it does not seem to be about EC abundances so I don't think it is extracting the pathway evidence features.
My guess would be that it should be something like:
ab_re_pp = self.nTotalComponents + self.nReacEvidenceFeatures + self.nPossibleClassFeatures
featureNonItem[ : , ab_re_pp + (idx * itemEvidenceFeaturesSize) : (idx * itemEvidenceFeaturesSize) + itemEvidenceFeaturesSize]