Advancing genome editing with artificial intelligence: opportunities, challenges, and future directions

Clustered regularly interspaced short palindromic repeat (CRISPR)-based genome editing (GED) technologies have unlocked exciting possibilities for understanding genes and improving medical treatments. On the other hand, Artificial intelligence (AI) helps genome editing achieve more precision, efficiency, and affordability in tackling various diseases, like Sickle cell anemia or Thalassemia. AI models have been in use for designing guide RNAs (gRNAs) for CRISPR-Cas systems. Tools like DeepCRISPR, CRISTA, and DeepHF have the capability to predict optimal guide RNAs (gRNAs) for a specified target sequence. These predictions take into account multiple factors, including genomic context, Cas protein type, desired mutation type, on-target/off-target scores, potential off-target sites, and the potential impacts of genome editing on gene function and cell phenotype. These models aid in optimizing different genome editing technologies, such as base, prime, and epigenome editing, which are advanced techniques to introduce precise and programmable changes to DNA sequences without relying on the homology-directed repair pathway or donor DNA templates. Furthermore, AI, in collaboration with genome editing and precision medicine, enables personalized treatments based on genetic profiles. AI analyzes patients' genomic data to identify mutations, variations, and biomarkers associated with different diseases like Cancer, Diabetes, Alzheimer's, etc. However, several challenges persist, including high costs, off-target editing, suitable delivery methods for CRISPR cargoes, improving editing efficiency, and ensuring safety in clinical applications. This review explores AI's contribution to improving CRISPR-based genome editing technologies and addresses existing challenges. It also discusses potential areas for future research in AI-driven CRISPR-based genome editing technologies. The integration of AI and genome editing opens up new possibilities for genetics, biomedicine, and healthcare, with significant implications for human health.

Artificial Intelligence/Machine Learning-Driven Small Molecule Repurposing via Off-Target Prediction and Transcriptomics

The process of discovering small molecule drugs involves screening numerous compounds and optimizing the most promising ones, both in vitro and in vivo. However, approximately 90% of these optimized candidates fail during trials due to unexpected toxicity or insufficient efficacy. Current concepts with respect to drug-protein interactions suggest that each small molecule interacts with an average of 6-11 targets. This implies that approved drugs and even discontinued compounds could be repurposed by leveraging their interactions with unintended targets. Therefore, we developed a computational repurposing framework for small molecules, which combines artificial intelligence/machine learning (AI/ML)-based and chemical similarity-based target prediction methods with cross-species transcriptomics information. This repurposing methodology incorporates eight distinct target prediction methods, including three machine learning methods. By using multiple orthogonal methods for a "dataset" composed of 2766 FDA-approved drugs targeting multiple therapeutic target classes, we identified 27,371 off-target interactions involving 2013 protein targets (i.e., an average of around 10 interactions per drug). Relative to the drugs in the dataset, we identified 150,620 structurally similar compounds. The highest number of predicted interactions were for drugs targeting G protein-coupled receptors (GPCRs), enzymes, and kinases with 10,648, 4081, and 3678 interactions, respectively. Notably, 17,283 (63%) of the off-target interactions have been confirmed in vitro. Approximately 4000 interactions had an IC50 of <100 nM for 1105 FDA-approved drugs and 1661 interactions had an IC50 of <10 nM for 696 FDA-approved drugs. Together, the confirmation of numerous predicted interactions and the exploration of tissue-specific expression patterns in human and animal tissues offer insights into potential drug repurposing for new therapeutic applications.

R-CRISPR: A Deep Learning Network to Predict Off-Target Activities with Mismatch, Insertion and Deletion in CRISPR-Cas9 System

The Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-associated protein 9 (Cas9) system is a groundbreaking gene-editing tool, which has been widely adopted in biomedical research. However, the guide RNAs in CRISPR-Cas9 system may induce unwanted off-target activities and further affect the practical application of the technique. Most existing in silico prediction methods that focused on off-target activities possess limited predictive precision and remain to be improved. Hence, it is necessary to propose a new in silico prediction method to address this problem. In this work, a deep learning framework named R-CRISPR is presented, which devises an encoding scheme to encode gRNA-target sequences into binary matrices, a convolutional neural network as feature extractor, and a recurrent neural network to predict off-target activities with mismatch, insertion, or deletion. It is demonstrated that R-CRISPR surpasses six mainstream prediction methods with a significant improvement on mismatch-only datasets verified by GUIDE-seq. Compared with the state-of-art prediction methods, R-CRISPR also achieves competitive performance on datasets with mismatch, insertion, and deletion. Furthermore, experiments show that data concatenate could influence the quality of training data, and investigate the optimal combination of datasets.

TCR Fingerprinting and Off-Target Peptide Identification

Adoptive T cell therapy using patient T cells redirected to recognize tumor-specific antigens by expressing genetically engineered high-affinity T-cell receptors (TCRs) has therapeutic potential for melanoma and other solid tumors. Clinical trials implementing genetically modified TCRs in melanoma patients have raised concerns regarding off-target toxicities resulting in lethal destruction of healthy tissue, highlighting the urgency of assessing which off-target peptides can be recognized by a TCR. As a model system we used the clinically efficacious NY-ESO-1-specific TCR C²⁵⁹, which recognizes the peptide epitope SLLMWITQC presented by HLA-A^*02:01. We investigated which amino acids at each position enable a TCR interaction by sequentially replacing every amino acid position outside of anchor positions 2 and 9 with all 19 possible alternative amino acids, resulting in 134 peptides (133 altered peptides plus epitope peptide). Each peptide was individually evaluated using three different in vitro assays: binding of the NY-ESO^c259 TCR to the peptide, peptide-dependent activation of TCR-expressing cells, and killing of peptide-presenting target cells. To represent the TCR recognition kernel, we defined Position Weight Matrices (PWMs) for each assay by assigning normalized measurements to each of the 20 amino acids in each position. To predict potential off-target peptides, we applied a novel algorithm projecting the PWM-defined kernel into the human proteome, scoring NY-ESO^c259 TCR recognition of 336,921 predicted human HLA-A^*02:01 binding 9-mer peptides. Of the 12 peptides with high predicted score, we confirmed 7 (including NY-ESO-1 antigen SLLMWITQC) strongly activate human primary NY-ESO^c259-expressing T cells. These off-target peptides include peptides with up to 7 amino acid changes (of 9 possible), which could not be predicted using the recognition motif as determined by alanine scans. Thus, this replacement scan assay determines the “TCR fingerprint” and, when coupled with the algorithm applied to the database of human 9-mer peptides binding to HLA-A^*02:01, enables the identification of potential off-target antigens and the tissues where they are expressed. This platform enables both screening of multiple TCRs to identify the best candidate for clinical development and identification of TCR-specific cross-reactive peptide recognition and constitutes an improved methodology for the identification of potential off-target peptides presented on MHC class I molecules.

Structural insight into the serotonin (5-HT) receptor family by molecular docking, molecular dynamics simulation and systems pharmacology analysis

Serotonin (5-HT) receptors are proteins involved in various neurological and biological processes, such as aggression, anxiety, appetite, cognition, learning, memory, mood, sleep, and thermoregulation. They are commonly associated with drug abuse and addiction due to their importance as targets for various pharmaceutical and recreational drugs. However, due to a high sequence similarity/identity among 5-HT receptors and the unavailability of the 3D structure of the different 5-HT receptor, no report was available so far regarding the systematical comparison of the key and selective residues involved in the binding pocket, making it difficult to design subtype-selective serotonergic drugs. In this work, we first built and validated three-dimensional models for all 5-HT receptors based on the existing crystal structures of 5-HT_1B, 5-HT_2B, and 5-HT_2C. Then, we performed molecular docking studies between 5-HT receptors agonists/inhibitors and our 3D models. The results from docking were consistent with the known binding affinities of each model. Sequentially, we compared the binding pose and selective residues among 5-HT receptors. Our results showed that the affinity variation could be potentially attributed to the selective residues located in the binding pockets. Moreover, we performed MD simulations for 12 5-HT receptors complexed with ligands; the results were consistent with our docking results and the reported data. Finally, we carried out off-target prediction and blood-brain barrier (BBB) prediction for Captagon using our established hallucinogen-related chemogenomics knowledgebase and in-house computational tools, with the hope to provide more information regarding the use of Captagon. We showed that 5-HT_2C, 5-HT_5A, and 5-HT₇ were the most promising targets for Captagon before metabolism. Overall, our findings can provide insights into future drug discovery and design of medications with high specificity to the individual 5-HT receptor to decrease the risk of addiction and prevent drug abuse.

Data imbalance in CRISPR off-target prediction

For genome-wide CRISPR off-target cleavage sites (OTS) prediction, an important issue is data imbalance-the number of true OTS recognized by whole-genome off-target detection techniques is much smaller than that of all possible nucleotide mismatch loci, making the training of machine learning model very challenging. Therefore, computational models proposed for OTS prediction and scoring should be carefully designed and properly evaluated in order to avoid bias. In our study, two tools are taken as examples to further emphasize the data imbalance issue in CRISPR off-target prediction to achieve better sensitivity and specificity for optimized CRISPR gene editing. We would like to indicate that (1) the benchmark of CRISPR off-target prediction should be properly evaluated and not overestimated by considering data imbalance issue; (2) incorporation of efficient computational techniques (including ensemble learning and data synthesis techniques) can help to address the data imbalance issue and improve the performance of CRISPR off-target prediction. Taking together, we call for more efforts to address the data imbalance issue in CRISPR off-target prediction to facilitate clinical utility of CRISPR-based gene editing techniques.

Hybridization Kinetics Explains CRISPR-Cas Off-Targeting Rules

Due to their specificity, efficiency, and ease of programming, CRISPR-associated nucleases are popular tools for genome editing. On the genomic scale, these nucleases still show considerable off-target activity though, posing a serious obstacle to the development of therapies. Off targeting is often minimized by choosing especially high-specificity guide sequences, based on algorithms that codify empirically determined off-targeting rules. A lack of mechanistic understanding of these rules has so far necessitated their ad hoc implementation, likely contributing to the limited precision of present algorithms. To understand the targeting rules, we kinetically model the physics of guide-target hybrid formation. Using only four parameters, our model elucidates the kinetic origin of the experimentally observed off-targeting rules, thereby rationalizing the results from both binding and cleavage assays. We favorably compare our model to published data from CRISPR-Cas9, CRISPR-Cpf1, CRISPR-Cascade, as well as the human Argonaute 2 system.