Will CRISPR-based therapy receive first FDA approval soon?

It has been more than a decade since the publication of the groundbreaking CRISPR paper in 2012. In spite of attention-grabbing controversies such as the patent dispute and the CRISPR baby scandal, CRISPR, along with many new nucleotide editing technologies, has brought about a paradigm shift in genetics, biotechnology, and drug development.

Today (October 31, 2023), an FDA advisory committee is set to review exagamglogene autotemcel (exa-cel) developed by Vertex Pharmaceuticals and CRISPR Therapeutics for sickle cell disease (SCD), which could become the first-ever CRISPR-based therapy approved by the FDA. The outcome of this committee discussion will play a pivotal role in FDA's final decision, which is expected due on December 8.

Data diversity is the key

FDA is mainly concerned about the off-target issue. SCD disproportionately affects individuals of African descent, but the analysis provided by Vertex and CRISPR Tx exclusively relied on the human reference genome and a mere 61 samples from African individuals from the 1000 Genome Project. FDA pointed out that at least one SNP, rs114518452, was missed in their analysis. The SNP was discovered using a more diverse dataset and could cause off-target editing.

This is a good lesson to learn, particularly for data science and AI applications in medicine, that it is crucial to account for data diversity and the potential disparities between training data and real-world data.

AI helps design the guide RNAs

Many machine learning methods were developed to predict the on- and off-target editing for CRISPR guide RNAs. A recent study conducted an extensive evaluation of various deep learning-based predictors, utilizing diverse datasets and sampling techniques. While most methods exhibited excellent performance on balanced datasets, there remains substantial room for improvement when confronted with moderately to severely imbalanced datasets.

In a noteworthy development, a method employing transfer learning for the prediction of prime-editing guide RNAs was featured in Nature Machine Intelligence last week. A transformer model was first trained on a large dataset but only comprising transversion editings from G to C at position +5, and then was fine-tuned on smaller datasets associated with diverse edit types and positions. Transfer learning is a typical strategy to deal with limited sample sizes.

AI helps increase the efficiency of pooled CRISPR screens

A novel technology, termed Compressed Perturb-seq, has been introduced to significantly reduce the requisite number of samples for conducting Perturb-seq experiments. Pertub-seq measures the transcriptional profiles from pooled CRISPR perturbations using single-cell RNA-Seq. Conventional Pertub-seq needs n cells to evaluate n perturbations. However, due to the sparsity inherent in regulatory circuits, multiple perturbations can be consolidated within a single cell, and the effect of individual perturbations can be computationally decomposed using matrix factorization. k*log(n) composite samples suffice to accurately recover the effects of n perturbations.

##########

If you find the newsletter helpful, please consider:

• 🔊 Sharing the newsletter with other people

• 👍 Upvoting on Product Hunt

• 📧 Sending any feedback, suggestions, and questions by directly replying to this email or writing reviews on Product Hunt

• 🙏 Supporting us with a cup of coffee.

Thanks, and see you next time!