14 September 2024 in Nucleic Acids Research
CRISPR-based genome editing technologies have revolutionised the field of molecular biology, offering unprecedented opportunities for precise genetic manipulation. However, off-target effects remain a significant challenge, potentially leading to unintended consequences and limiting the applicability of CRISPR-based genome editing technologies in clinical settings. Current literature predominantly focuses on point predictions for off-target activity, which may not fully capture the range of possible outcomes and associated risks. Here, we present crispAI, a neural network architecture-based approach for predicting uncertainty estimates for off-target cleavage activity, providing a more comprehensive risk assessment and facilitating improved decision-making in single guide RNA (sgRNA) design. Our approach makes use of the count noise model Zero Inflated Negative Binomial (ZINB) to model the uncertainty in the off-target cleavage activity data. In addition, we present the first-of-its-kind genome-wide sgRNA efficiency score, crispAI-aggregate, enabling prioritization among sgRNAs with similar point aggregate predictions by providing richer information compared to existing aggregate scores. We show that uncertainty estimates of our approach are calibrated and its predictive performance is superior to the state-of-the-art in silico off-target cleavage activity prediction methods. The tool and the trained models are available at https://github.com/furkanozdenn/crispr-offtarget-uncertainty.
13 June 2022 in Genome Research
Accurate and efficient detection of copy number variants (CNVs) is of critical importance owing to their significant association with complex genetic diseases. Although algorithms that use whole-genome sequencing (WGS) data provide stable results with mostly valid statistical assumptions, copy number detection on whole-exome sequencing (WES) data shows comparatively lower accuracy. This is unfortunate as WES data are cost-efficient, compact, and relatively ubiquitous. The bottleneck is primarily due to the noncontiguous nature of the targeted capture: biases in targeted genomic hybridization, GC content, targeting probes, and sample batching during sequencing. Here, we present a novel deep learning model, DECoNT, which uses the matched WES and WGS data, and learns to correct the copy number variations reported by any off-the-shelf WES-based germline CNV caller. We train DECoNT on the 1000 Genomes Project data, and we show that we can efficiently triple the duplication call precision and double the deletion call precision of the state-of-the-art algorithms. We also show that our model consistently improves the performance independent of (1) sequencing technology, (2) exome capture kit, and (3) CNV caller. Using DECoNT as a universal exome CNV call polisher has the potential to improve the reliability of germline CNV detection on WES data sets.
18 February 2026 in medRxiv
Diagnosing rare genetic disease hinges on identifying the causal variant among thousands of candidates, a task that leaves most of the 300 million affected patients undiagnosed despite the availability of genome sequencing. Here we present DAVP (Deep Agentic Variant Prioritisation), a hierarchical agentic-LLM pipeline combining gene pre-filtering, knowledge-graph-grounded variant reports, and iterative LLM tournaments to rank variants against each patient’s phenotype. We benchmark DAVP on 709 diagnosis cases across four datasets, including LIRDB-47—a new cohort of 47 Turkish rare-disease patients with 63 clinically diagnosed causal variants. DAVP achieves 85.5% variant-level top-3 recall on the full benchmark; on LIRDB-47, it places the causal variant in the top-3 in 76.2% of patients (top-20 in 88.9%), a 1.6-fold improvement over the best phenotype-driven baseline (Exomiser, 47.6%). Our work suggests that agentic LLM systems combining evidence synthesis with patient-specific reasoning have the potential to reshape clinical genomic workflows.
10 December 2025 in bioRxiv
Prime editing has emerged as a versatile genome editing, technology capable of installing precise genetic modifications without requiring double-strand breaks or donor templates. However, designing pegRNAs with high editing efficiency remains a challenge due to the complex interplay of sequence features that affect editing outcomes. Current approaches predominantly provide point predictions without capturing the inherent uncertainty in editing efficiency, limiting risk assessment, and decision-making in pegRNA design. Here, we present crispAIPE, a transformer-based probabilistic framework for predicting prime editing variant correction efficiencywith uncertainty quantification. Our approach models editing outcomes in a 3D simplex space, enabling comprehensive uncertainty estimation while achieving superior predictive performance compared to existing models. crispAIPE leverages transformer encoders to capture long-range sequence dependencies and contextual relationships, surpassing existing models in the point estimate prediction task. The model also predicts efficiency distributions for all edit types, including single nucleotide replacements, insertions, and deletions. Trained on 73, 939 pegRNAs in multiple cell lines, for all outcome types, on overall, crispAIPE achieves a Spearman correlation of 0.881 and a Pearson correlation of 0.894 while providing calibrated uncertainty estimates that allow the selection of risk-sensitive pegRNA. Additionally, we identify key sequence motifs and positional features that drive editing efficiency, providing interpretable insights into the sequence determinants of prime editing. We demonstrate that uncertainty-aware predictions could significantly improve pegRNA design outcomes, with high-confidence predictions showing higher success rates compared to low-confidence designs. crispAIPE represents the first probabilistic deep learning framework for prime editing, bridging the gap between predictive accuracy and uncertainty quantification to enable more reliable and interpretable pegRNA design. The source code and example data are available at https://github.com/furkanozdenn/pe-uncert.