Net-Net Auto Machine Learning (AutoML) Prediction of Complex Ecosystems.

Area of Science:

• Ecology
• Computational Biology
• Machine Learning

Background:

• Biological Ecosystem Networks (BENs) represent complex species interactions, but experimental validation of all trophic links is infeasible.
• Computational prediction of BENs is crucial due to the vast amount of data generated by experimental methods.
• Artificial Neural Networks (ANNs) offer a machine learning approach for BEN prediction, but selecting optimal ANN topologies remains difficult.

Purpose of the Study:

• To introduce and evaluate the novel Net-Net AutoML approach for automated selection of Artificial Neural Networks (ANNs) for Biological Ecosystem Network (BEN) prediction.
• To investigate the efficacy of using Shannon entropy (Shk) measures from both BENs and ANN topologies for optimizing model selection.
• To compare the performance of twelve different classifier types within the Net-Net AutoML framework.

Main Methods:

• Developed the Net-Net AutoML approach, utilizing Shannon entropy (Shk) values of BENs and ANN topologies.
• Trained and tested twelve distinct classifier types, including linear models, Bayesian methods, tree-based algorithms, multilayer perceptrons, and deep neural networks.
• Evaluated models on 69 BENs, predicting 338,050 potential links across 10 ANN topologies.

Main Results:

• The best performing Net-Net AutoML model utilized a deep fully connected neural network.
• This optimal model achieved a test accuracy of 0.866 and a test AUROC of 0.935.
• The study demonstrated the effectiveness of AutoML in selecting high-accuracy ANNs for ecological network prediction.

Conclusions:

• The Net-Net AutoML approach successfully automates the selection of efficient ANNs for predicting Biological Ecosystem Networks.
• Deep fully connected neural networks proved most effective within this AutoML framework for ecological link prediction.
• This methodology offers a promising pathway for applying AutoML to diverse systems and machine learning algorithms in computational biology.