Over the last decade, next generation sequencing has become widely implemented in clinical practice. However, as genetic variants of uncertain significance (VUS) are frequently identified, the need for scaled functional interpretation of such variants has become increasingly apparent. One method to address this is saturation genome editing (SGE), which allows for scaled multiplexed functional assessment of single nucleotide variants. The current applications of SGE, however, rely on homology-directed repair (HDR) to introduce variants of interest, which is limited by low editing efficiencies and low product purity. Here, we have adapted CRISPR prime editing for SGE and demonstrated its utility in understanding the functional significance of variants in the NPC1 gene underlying the lysosomal storage disorder Niemann-Pick disease type C1 (NPC). Additionally, we have designed a genome editing strategy that allows for the haploidization of gene loci, which permits isolated variant interpretation in virtually any cell type. By combining saturation prime editing (SPE) with a clinically relevant assay, we have functionally scored and interpreted 256 variants in NPC1 haploidized HEK293T cells. To further demonstrate the applicability of this strategy, we used SPE and cell model haploidization to functionally score 465 variants in the BRCA2 gene. We anticipate that our work will be translatable to any gene with an appropriate cellular assay, allowing for more rapid and accurate diagnosis and improved genetic counselling and ultimately precise patient care.

Image noise is a common problem in light microscopy. This is particularly true in real-time live-cell imaging applications in which long-term cell viability necessitates low-light conditions. Modern denoisers are typically trained on a representative dataset, sometimes consisting of just unpaired noisy shots. However, when data are acquired in real time to track dynamic cellular processes, it is not always practical nor economical to generate these training sets. Recently, denoisers have emerged that allow us to denoise single images without a training set or knowledge about the underlying noise. But such methods are currently too slow to be integrated into imaging pipelines that require rapid, real-time hardware feedback. Here we present Noise2Fast, which can overcome these limitations. Noise2Fast uses a novel downsampling technique we refer to as ‘chequerboard downsampling’. This allows us to train on a discrete 4-image training set, while convergence can be monitored using the original noisy image. We show that Noise2Fast is faster than all similar methods with only a small drop in accuracy compared to the gold standard. We integrate Noise2Fast into real-time multi-modal imaging applications and demonstrate its broad applicability to diverse imaging and analysis pipelines.