False Positives and Negative Results in Protein Sciences

Frontiers in Biotechnology and Bioengineering is launching a special issue titled – "False Positives and Negative Results in Protein Sciences" within the Industrial Biotechnology section is calling for research articles exploring the interpretability of (a) false positive predictions, and (b) negative results in protein science relevant to industrial biotechnology. We invite submissions that shed light on the often opaque nature of machine learning (ML), optimization algorithms, and theoretically-but-rationally driven structural protein biology efforts.

Proteins are the workhorses of the cellular world, and understanding their structure, function, interactions, and dynamics is crucial for industrial biotechnology advancements. However, the surge in computational tools, particularly ML and optimization programs, has brought both promise and challenges. While these methods have accelerated protein research, very little literature exists that investigate, in detail, the cases when these experiments do not line up with experimental truth. Notwithstanding, gleaning such insights is key to annotating utility and appropriateness of different prediction platforms for different tasks. False positives, where a method predicts a function or interaction that doesn't exist, and negative results that overlook potential leads, hinder progress and waste valuable resources in industrial pipelines.

We seek to bridge this gap by publishing high-quality research that prioritizes explainable artificial intelligence (XAI), optimization (XOpt), and intuition building in protein science. We are particularly interested in manuscripts that:

(a) Decipher False Positives – explore the reasons behind false positive structure and function prediction, identify biases in training data, limitations of specific algorithms, or inherent difficulties in modeling complex protein behavior. Develop strategies for validating and filtering ML outputs to avoid misleading conclusions.

(b) Unravel Negative Results– delve into the factors contributing to negative results in protein engineering or interaction studies, evaluate the effectiveness of current optimization algorithms and design hypotheses, and propose methods for dissecting negative results to identify potentially overlooked interactions or unexplored design avenues.

(c) Explainable AI and Protein Science– embrace Explainable Artificial Intelligence (XAI) in the context of protein research, develop methods to understand the decision-making processes within ML models used for protein structure prediction or function assignment, and utilize visualization tools and feature importance analysis to translate complex model outputs into biologically meaningful insights.

(d) Bridging Theory and Experiment– explore the limitations of rational design approaches based on sequence or structural information in predicting protein properties, investigate strategies to integrate experimental data, and computational modeling to refine hypotheses.

We encourage authors to design such negative datasets, predictions, and results in these studies and make them available to the research community through long-term data hosting services such as (not limited to) GitHub and HuggingFace.