DTMP-Prime Model Advances Prime Editing Predictions

DTMP-Prime combines sophisticated feature selection, deep embedding, and multi-head attention layers to predict the efficiency of PE complexes by evaluating factors affecting pegRNA and target DNA sequence interactions. More than 43 features are used to enable the detailed analysis that reduces off-target effects and enhances editing outcomes, including pegRNA length, RNA folding structures, Cas9 activity, and nucleotide composition.

DTMP-Prime also uses DNABERT embeddings to capture nuanced relationships within DNA sequences, effectively identifying optimal pegRNA configurations for various PE systems, like PE2 and PE3. Performance comparisons with state-of-the-art PE prediction tools, such as DeepPE and Easy-Prime, demonstrate that DTMP-Prime achieves higher precision, with Pearson and Spearman correlation coefficients of 0.8 and 0.77, respectively, in test datasets.

This study was led by Leila Safari from the University of Zanjan, Iran, and Alireza Khanteymoori from the University of Freiburg. It was published today in Molecular Therapy: Nucleic Acids.