Enzyme mechanism prediction: a template matching problem on InterPro signature subspaces

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Previous work demonstrated high accuracy in predicting enzyme chemical mechanisms using a k Nearest Neighbour (kNN) rule with InterPro sequence signatures.
• The kNN algorithm's sensitivity to training data errors and small datasets was a concern in prior studies.
• The current study re-analyzed the dataset and prediction results to understand the observed high classification performance.

Purpose of the Study:

• To explain the remarkable classification performance of a k Nearest Neighbour (kNN) rule for predicting enzyme chemical mechanisms.
• To investigate the underlying reasons for the high accuracy despite a large and sparse feature space.

Main Methods:

• Utilized a k Nearest Neighbour (kNN) rule with k=1 (k1NN).
• Employed 321 InterPro sequence signatures as enzyme features.
• Re-analyzed a dataset of 248 enzymes with 71 MACiE database mechanism labels.

Main Results:

• Enzymes with different chemical mechanisms were found in barely overlapping subspaces within the feature space.
• The selected InterPro features contained sufficient information for accurate enzymatic mechanism classification.
• The classification problem was effectively reduced to a simple look-up exercise due to feature distinctiveness.

Conclusions:

• The study explains the "anomaly" of a basic kNN algorithm achieving high performance in a large, sparse feature space for enzyme mechanism prediction.
• InterPro signatures were confirmed as critical for accurate enzyme mechanism prediction.
• Simple rules were suggested for inductively predicting predefined mechanisms in novel enzymes.