[Paper Review] Antifragility Predicts the Robustness and Evolvability of Biological Networks through Multi-class Classification with a Convolutional Neural Network

Hyobin Kim, Stalin Muñoz+2 more|arXiv (Cornell University)|Feb 4, 2020

Bioinformatics and Genomic Networks72 references4 citations

TL;DR

This paper proposes a novel method to predict the robustness and evolvability of biological networks using antifragility as a predictor, leveraging a convolutional neural network (CNN) to classify network properties based on differences in antifragility between original and mutated Boolean network models. The CNN achieves high accuracy in multi-class classification, demonstrating that antifragility is a strong predictor of network resilience and evolvability without requiring explicit functional comparison after mutation.

ABSTRACT

Robustness and evolvability are essential properties to the evolution of biological networks. To determine if a biological network is robust and/or evolvable, it is required to compare its functions before and after mutations. However, this sometimes takes a high computational cost as the network size grows. Here we develop a predictive method to estimate the robustness and evolvability of biological networks without an explicit comparison of functions. We measure antifragility in Boolean network models of biological systems and use this as the predictor. Antifragility occurs when a system benefits from external perturbations. By means of the differences of antifragility between the original and mutated biological networks, we train a convolutional neural network (CNN) and test it to classify the properties of robustness and evolvability. We found that our CNN model successfully classified the properties. Thus, we conclude that our antifragility measure can be used as a predictor of the robustness and evolvability of biological networks.

Motivation & Objective

• To address the high computational cost of evaluating robustness and evolvability in biological networks through explicit functional comparison after mutation.
• To investigate whether antifragility—defined as a system's benefit from external perturbations—can serve as a predictive measure for network resilience and adaptability.
• To develop a machine learning framework that classifies network properties (robustness and evolvability) using antifragility differences between original and mutated networks.
• To reduce reliance on computationally expensive simulations by replacing direct functional comparison with a learned predictive model.

Proposed method

• Model biological networks using Boolean network dynamics to simulate regulatory interactions.
• Quantify antifragility in both original and mutated network configurations using statistical measures of response to perturbations.
• Compute the difference in antifragility between original and mutated networks as input features for classification.
• Train a convolutional neural network (CNN) on these antifragility difference features to perform multi-class classification of network properties.
• Use the CNN to predict whether a network is robust, evolvable, both, or neither, based on its antifragility profile.
• Validate the model using cross-validation and assess performance via classification accuracy and F1-score on test sets.

Experimental results

Research questions

• RQ1 Can antifragility serve as a reliable predictor of robustness and evolvability in biological networks?
• RQ2 How accurately can a convolutional neural network classify the robustness and evolvability of biological networks using antifragility differences as input?
• RQ3 Does the proposed method reduce computational cost compared to traditional approaches that require explicit functional comparison after mutation?
• RQ4 What is the relative contribution of antifragility to predicting network resilience and adaptability in complex regulatory systems?

Key findings

• The convolutional neural network achieved high classification accuracy in predicting the combined properties of robustness and evolvability in biological networks.
• Antifragility differences between original and mutated networks were found to be a strong and informative predictor of network behavior.
• The model successfully classified networks into four distinct classes: robust and evolvable, robust but not evolvable, not robust but evolvable, and neither robust nor evolvable.
• The approach significantly reduces the need for computationally intensive functional comparisons after mutation, offering a scalable alternative for systems biology.
• The results demonstrate that antifragility is not only a measure of system resilience but also a key indicator of evolutionary potential in regulatory networks.
• The method was validated on Boolean network models and published in Entropy, with a DOI linking to the final peer-reviewed version.