Reducing DNA/RNA Sequence Dropout with Predictive Synthesis Models

Stanford researchers have developed a method using trained learning models to optimize synthetic DNA libraries for high-throughput molecular biology experiments. This approach ensures more uniform sequence coverage by using predictive models to avoid sequence "dropout" caused by synthesis and downstream workflows.

High-throughput experiments, such as CRISPR screens and synthetic RNA/DNA testing, often face challenges where many sequences are lost during the final sequencing readout. This occurs because sequences with higher read potential can dominate those with lower potential, reducing overall efficiency and leading to financial waste. Stanford scientists have developed predictive models, or "oracles," to predict sequence performance and split oligonucleotide pools into subpools.

The method applies to a wide range of biological research areas, including CRISPR screens, synthetic biology, gene synthesis, DNA-based storage, and reporter systems. Researchers can guide the subpool-splitting process using pilot experiments or AI-driven predictions tailored to specific workflows.

Figure