Integrating Deep Learning and Synthetic Biology: A\nCo-Design Approach for Enhancing Gene Expression via N‑Terminal\nCoding Sequences

N-terminal\ncoding sequence (NCS) influences gene expression by\nimpacting the translation initiation rate. The NCS optimization problem\nis to find an NCS that maximizes gene expression. The problem is important\nin genetic engineering. However, current methods for NCS optimization\nsuch as rational design and statistics-guided approaches are labor-intensive\nyield only relatively small improvements. This paper introduces a\ndeep learning/synthetic biology codesigned few-shot training workflow\nfor NCS optimization. Our method utilizes <i>k</i>-nearest\nencoding followed by word2vec to encode the NCS, then performs feature\nextraction using attention mechanisms, before constructing a time-series\nnetwork for predicting gene expression intensity, and finally a direct\nsearch algorithm identifies the optimal NCS with limited training\ndata. We took green fluorescent protein (GFP) expressed by Bacillus subtilis as a reporting protein of NCSs,\nand employed the fluorescence enhancement factor as the metric of\nNCS optimization. Within just six iterative experiments, our model\ngenerated an NCS (MLD<sub>62</sub>) that increased average GFP expression\nby 5.41-fold, outperforming the state-of-the-art NCS designs. Extending\nour findings beyond GFP, we showed that our engineered NCS (MLD<sub>62</sub>) can effectively boost the production of N-acetylneuraminic\nacid by enhancing the expression of the crucial rate-limiting <i>GNA1</i> gene, demonstrating its practical utility. We have\nopen-sourced our NCS expression database and experimental procedures\nfor public use.