Screen2Learn - Exploiter la biodiversité : une approche de criblage et d'apprentissage

An appealing way to improve lignocellulosic enzyme cocktails, mostly produced by the filamentous fungus Trichoderma reesei is its complementation with new enzymes retrieved from biodiversity. In order to select efficient enzymes during the screening phase, a good expression and secretion by a convenient and easily transformable host organism is necessary for every protein to be tested. However, the success of heterologous protein production in eukaryotic organisms, such as S. cerevisiae or P. pastoris is unpredictable and depends in unclear ways on the protein to produce. To predict and select the best expression strategy, the impact of parameters such as preferential codon usage, peptide signal and factors involved in the secretion system have to be evaluated. Libraries of 105 strains with different genetic backgrounds will be constructed for 36 enzymes and screened by the recently developed “auto-growth” method, based on the coupling of gowth rate and secretability. Efficient secretion strains will grow faster thus allowing to identify the best combinations by deep sequencing. In a second step, a machine learning approach will be used in order to develop predictive models of the impact of secretion conditions for a given protein. Finally, the 36 enzymes will be produced in optimal and sub-optimal conditions to test the predictive power of the model, and their activity tested.