SoluProt: Prediction of Protein Solubility

SoluProt: Prediction of Protein Solubility

Paper in proceedings outside WoS and Scopus

Protein solubility poses a major bottleneck in productionof many therapeutic and industrially attractive proteins. Experimentalsolubilization attempts are plagued by relatively low success rates andoften lead to the loss of biological activity. Therefore, any advance incomputational prediction of protein solubility may reduce the cost of experimentalstudies significantly. Here, we propose a novel software toolSoluProt for prediction of solubility from protein sequence based on machinelearning and TargetTrack database. SoluProt achieved the bestaccuracy 58.2% and AUC 0.61 of all available tools at an independentbalanced test set derived from NESG database. While the absolute predictionperformance is rather low, SoluProt can still help to reduce costsof experimental studies significantly by efficient prioritization of proteinsequences. The main SoluProt contribution lies in improved preprocessingof noisy training data and sensible selection of sequence featuresincluded in the prediction model.

Protein solubility poses a major bottleneck in productionof many therapeutic and industrially attractive proteins. Experimentalsolubilization attempts are plagued by relatively low success rates andoften lead to the loss of biological activity. Therefore, any advance incomputational prediction of protein solubility may reduce the cost of experimentalstudies significantly. Here, we propose a novel software toolSoluProt for prediction of solubility from protein sequence based on machinelearning and TargetTrack database. SoluProt achieved the bestaccuracy 58.2% and AUC 0.61 of all available tools at an independentbalanced test set derived from NESG database. While the absolute predictionperformance is rather low, SoluProt can still help to reduce costsof experimental studies significantly by efficient prioritization of proteinsequences. The main SoluProt contribution lies in improved preprocessingof noisy training data and sensible selection of sequence featuresincluded in the prediction model.

protein, solubility, prediction, machine-learning

protein, solubility, prediction, machine-learning

HON, J.; MARUŠIAK, M.; MARTÍNEK, T.; ZENDULKA, J.; BEDNÁŘ, D.; DAMBORSKÝ, J.

2019

17.08.2018

Brno University of Technology

Brno

978-80-214-5679-2

DAZ & WIKT 2018 Proceedings

261

265

5

https://www.fit.vut.cz/research/publication/11808/

DAZ_S9_Jiri_Hon_v2