# ProteoAutoNet: high-throughput co-eluted protein analysis with robotics and machine learning

## TL;DR

ProteoAutoNet is a robotic and machine learning system that improves the study of protein interactions in cancer cells, revealing new disease-related protein complexes.

## Contribution

ProteoAutoNet combines robotics and machine learning to increase CF-MS throughput and predict thousands of co-eluted proteins in thyroid cancer cells.

## Key findings

- ProteoAutoNet doubles the throughput of CF-MS sample processing.
- The model predicted 25,173 co-eluted proteins with an AUROC of 0.78 in thyroid cell lines.
- Upregulated proteasome, prefoldin complexes, and a TGM2-HK1 interaction were identified in thyroid cancer cells.

## Abstract

Co-fractionation mass spectrometry (CF-MS) enables large-scale profiling of endogenous protein-protein interactions, yet CF-MS data generation is of low throughput and therefore predictive models are often limited by the scarcity and limited diversity of high-quality training data. To address this, we present ProteoAutoNet, a robotic experimental platform integrated with a computational workflow for high-throughput CF-MS analysis. This workflow increases the throughput of sample processing from protein complex to peptide by about two times. The integrated machine learning model incorporates targeted data augmentation to expand and diversify reliable protein interaction data, thereby improving model robustness. When applied to three thyroid cell lines, the model predicted 25,173 co-eluted proteins with an AUROC of 0.78. This analysis revealed significantly upregulated proteasome and prefoldin complexes in the lung metastatic follicular thyroid carcinoma cell line FTC238 compared with the normal thyroid cell line Nthy-ori 3-1. Notably, we identified a protein interaction between TGM2 and HK1 that was significantly upregulated in the papillary thyroid carcinoma cell line TPC-1. ProteoAutoNet provides an improved framework for investigating protein-protein interactions and uncovering interactions.

Co-fractionation mass spectrometry reveals protein interactions but is low throughput and data-limited. Here, the authors present ProteoAutoNet, a robotic platform integrated with machine learning that doubles sample processing throughput and predicts thousands of co-eluted protein interactions in thyroid cancer cells, revealing disease-relevant complexes like upregulated proteasome, prefoldin, and a predicted HK1-TGM2 interaction.