Can We Quickly Learn to "Translate" Bioactive Molecules with Transformer Models?

Meaningful exploration of the chemical space of druglike molecules in drug design is a highly challenging task due to a combinatorial explosion of possible modifications of molecules. In this work, we address this problem with transformer models, a type of machine learning (ML) model originally developed for machine translation. By training transformer models on pairs of similar bioactive molecules from the public ChEMBL data set, we enable them to learn medicinal-chemistry-meaningful, context-dependent transformations of molecules, including those absent from the training set. By retrospective analysis on the performance of transformer models on ChEMBL subsets of ligands binding to COX2, DRD2, or HERG protein targets, we demonstrate that the models can generate structures identical or highly similar to most active ligands, despite the models having not seen any ligands active against the corresponding protein target during training. Our work demonstrates that human experts working on hit expansion in drug design can easily and quickly employ transformer models, originally developed to translate texts from one natural language to another, to "translate" from known molecules active against a given protein target to novel molecules active against the same target.

Directed differentiation of human iPSCs to functional ovarian granulosa-like cells via transcription factor overexpression.

An in vitro model of human ovarian follicles would greatly benefit the study of female reproduction. Ovarian development requires the combination of germ cells and several types of somatic cells. Among these, granulosa cells play a key role in follicle formation and support for oogenesis. Whereas efficient protocols exist for generating human primordial germ cell-like cells (hPGCLCs) from human induced pluripotent stem cells (hiPSCs), a method of generating granulosa cells has been elusive. Here, we report that simultaneous overexpression of two transcription factors (TFs) can direct the differentiation of hiPSCs to granulosa-like cells. We elucidate the regulatory effects of several granulosa-related TFs and establish that overexpression of NR5A1 and either RUNX1 or RUNX2 is sufficient to generate granulosa-like cells. Our granulosa-like cells have transcriptomes similar to human fetal ovarian cells and recapitulate key ovarian phenotypes including follicle formation and steroidogenesis. When aggregated with hPGCLCs, our cells form ovary-like organoids (ovaroids) and support hPGCLC development from the premigratory to the gonadal stage as measured by induction of DAZL expression. This model system will provide unique opportunities for studying human ovarian biology and may enable the development of therapies for female reproductive health.

Ovaries are responsible for forming the eggs humans and other mammals need to reproduce. Once mature, the egg cell is released into the fallopian tube where it can be potentially fertilized by a sperm. Despite their crucial role, how eggs are made in the ovary is poorly understood. This is because ovaries are hard to access, making it difficult to conduct experiments on them. To overcome this, researchers have built artificial ovaries in the laboratory using stem cells from the embryos of mice which can develop into all cell types in the adult body. By culturing these embryonic stem cells under special conditions, researchers can convert them in to the two main cell types of the developing ovary: germ cells which go on to form eggs, and granulosa cells which help eggs grow and mature. The resulting lab-grown ovary can make eggs that produce live mice when fertilized. This approach has also been applied to human induced pluripotent stem cells (iPSCs), adult human cells which have been reprogrammed to a stem-like state. While this has produced human germ cells, generating human granulosa cells has been more challenging. Here, Pierson Smela, Kramme et al. show that activating a specific set of transcription factors (proteins that switch genes on or off) in iPSCs can make them transition to granulosa cells. First, the team tested random combinations of 35 transcription factors which, based on previous literature and genetic data, were likely to play a role in the formation of granulosa cells. This led to the identification of a small number of factors that caused the human iPSCs to develop features and carry out roles seen in mature granulosa cells; this includes producing an important reproductive hormone and supporting the maturation of germ cells. Pierson Smela, Kramme et al. found that growing these granulosa-like cells together with germ cells (also generated via iPSCs) resulted in structures similar to ovarian follicles which help eggs develop. These findings could help researchers build stable systems for studying how granulosa cells behave in human ovaries. This could lead to new insights about reproductive health.

PAM-Flexible Genome Editing with an Engineered Chimeric Cas9.

CRISPR enzymes require a defined protospacer adjacent motif (PAM) flanking a guide RNA-programmed target site, limiting their sequence accessibility for robust genome editing applications. In this study, we recombine the PAM-interacting domain of SpRY, a broad-targeting Cas9 possessing an NRN > NYN PAM preference, with the N-terminus of Sc++, a Cas9 with simultaneously broad, efficient, and accurate NNG editing capabilities, to generate a chimeric enzyme with highly flexible PAM preference: SpRYc. We demonstrate that SpRYc leverages properties of both enzymes to specifically edit diverse NNN PAMs and disease-related loci for potential therapeutic applications. In total, the unique approaches to generate SpRYc, coupled with its robust flexibility, highlight the power of integrative protein design for Cas9 engineering and motivate downstream editing applications that require precise genomic positioning.

PAM-flexible genome editing with an engineered chimeric Cas9.

CRISPR enzymes require a defined protospacer adjacent motif (PAM) flanking a guide RNA-programmed target site, limiting their sequence accessibility for robust genome editing applications. In this study, we recombine the PAM-interacting domain of SpRY, a broad-targeting Cas9 possessing an NRN > NYN (R = A or G, Y = C or T) PAM preference, with the N-terminus of Sc + +, a Cas9 with simultaneously broad, efficient, and accurate NNG editing capabilities, to generate a chimeric enzyme with highly flexible PAM preference: SpRYc. We demonstrate that SpRYc leverages properties of both enzymes to specifically edit diverse PAMs and disease-related loci for potential therapeutic applications. In total, the approaches to generate SpRYc, coupled with its robust flexibility, highlight the power of integrative protein design for Cas9 engineering and motivate downstream editing applications that require precise genomic positioning.

A Cas9 with PAM recognition for adenine dinucleotides.

CRISPR-associated (Cas) DNA-endonucleases are remarkably effective tools for genome engineering, but have limited target ranges due to their protospacer adjacent motif (PAM) requirements. We demonstrate a critical expansion of the targetable sequence space for a type II-A CRISPR-associated enzyme through identification of the natural 5[Formula: see text]-NAAN-3[Formula: see text] PAM preference of Streptococcus macacae Cas9 (SmacCas9). To achieve efficient editing activity, we graft the PAM-interacting domain of SmacCas9 to its well-established ortholog from Streptococcus pyogenes (SpyCas9), and further engineer an increased efficiency variant (iSpyMac) for robust genome editing activity. We establish that our hybrids can target all adenine dinucleotide PAM sequences and possess robust and accurate editing capabilities in human cells.

An engineered ScCas9 with broad PAM range and high specificity and activity.

CRISPR enzymes require a protospacer-adjacent motif (PAM) near the target cleavage site, constraining the sequences accessible for editing. In the present study, we combine protein motifs from several orthologs to engineer two variants of Streptococcus canis Cas9-Sc++ and a higher-fidelity mutant HiFi-Sc++-that have simultaneously broad 5'-NNG-3' PAM compatibility, robust DNA-cleavage activity and minimal off-target activity. Sc++ and HiFi-Sc++ extend the use of CRISPR editing for diverse applications.

Publisher Correction: An engineered ScCas9 with broad PAM range and high specificity and activity.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.