Selection of target-binding proteins from the information of weakly enriched phage display libraries by deep sequencing and machine learning

Abstract

Despite the advances in surface-display systems for directed evolution, variants with high affinity are not always enriched due to undesirable biases that increase target-unrelated variants during biopanning. Here, our goal was to design a library containing improved variants from the information of the “weakly enriched” library where functional variants were weakly enriched. Deep sequencing for the previous biopanning result, where no functional antibody mimetics were experimentally identified, revealed that weak enrichment was partly due to undesirable biases during phage infection and amplification steps. The clustering analysis of the deep sequencing data from appropriate steps revealed no distinct sequence patterns, but a Bayesian machine learning model trained with the selected deep sequencing data supplied nine clusters with distinct sequence patterns. Phage libraries were designed on the basis of the sequence patterns identified, and four improved variants with target-specific affinity (EC 50 = 80–277 nM) were identified by biopanning. The selection and use of deep sequencing data without undesirable bias enabled us to extract the information on prospective variants. In summary, the use of appropriate deep sequencing data and machine learning with the sequence data has the possibility of finding sequence space where functional variants are enriched.

Development of a new affinity maturation protocol for the construction of an internalizing anti-nucleolin antibody library

“…However, the use of several stringent conditions (FACS to select a narrow window of events with clear increased fluorescence signal, PBS supplemented with 2% FBS as blocking solution, washing between incubations at 4 °C and 37 °C, and acidic stripping of cell surface-bound phages, that were not internalized after incubation at 37 °C) decrease the potential of collection of phages internalized in a non-specific manner. Together with reports that only one round of panning may not be sufficient to decrease the percentage of non-functional sequences, such as sequences with stop codons 68 , it is suggested that such value is due to the quality of the constructed library, sequencing errors 69 , or bias during bacteria infection 49 .…”

DSMBind: SE(3) denoising score matching for unsupervised binding energy prediction and nanobody design

“…DSMBind has several advantages over previous supervised learning approaches for binding prediction [11, 13, 26]. First, it does not rely on binding data collected from phage-display libraries, which is time-consuming and labor-intensive.…”

“…To study new types of protein interactions, we need to generate binding data through phage display [26] or ribosome display [4], which is time-consuming and labor-intensive. Models trained on such data data [11, 13, 26] are generally not applicable to other protein targets because its training data is collected for one specific protein. Ideally, we want an unsupervised learning approach that generalizes across protein and does not depend on expensive binder screening procedure.…”

Phage Display Technology in Biomarker Identification with Emphasis on Non-Cancerous Diseases

“…Utilizing deep sequencing data [150] and machine learning [151] can help identify sequence spaces containing functional variants enriched in deep sequencing data. Consequently, this approach aids in recognizing, categorizing, and quantifying features relevant to diagnostics and therapy [150][151][152].…”