RESUMEN
Viruses have evolved the ability to bind and enter cells through interactions with a wide variety of cell macromolecules. We engineered peptide-modified adeno-associated virus (AAV) capsids that transduce the brain through the introduction of de novo interactions with 2 proteins expressed on the mouse blood-brain barrier (BBB), LY6A or LY6C1. The in vivo tropisms of these capsids are predictable as they are dependent on the cell- and strain-specific expression of their target protein. This approach generated hundreds of capsids with dramatically enhanced central nervous system (CNS) tropisms within a single round of screening in vitro and secondary validation in vivo thereby reducing the use of animals in comparison to conventional multi-round in vivo selections. The reproducible and quantitative data derived via this method enabled both saturation mutagenesis and machine learning (ML)-guided exploration of the capsid sequence space. Notably, during our validation process, we determined that nearly all published AAV capsids that were selected for their ability to cross the BBB in mice leverage either the LY6A or LY6C1 protein, which are not present in primates. This work demonstrates that AAV capsids can be directly targeted to specific proteins to generate potent gene delivery vectors with known mechanisms of action and predictable tropisms.
Asunto(s)
Barrera Hematoencefálica , Cápside , Ratones , Animales , Barrera Hematoencefálica/metabolismo , Cápside/metabolismo , Vectores Genéticos , Sistema Nervioso Central/metabolismo , Proteínas de la Cápside/genética , Proteínas de la Cápside/metabolismo , Dependovirus/genética , Dependovirus/metabolismoRESUMEN
MOTIVATION: Can we predict protein-protein interactions (PPIs) of a novel virus with its host? Three major problems arise: the lack of known PPIs for that virus to learn from, the cost of learning about its proteins and the sequence dissimilarity among viral families that makes most methods inapplicable or inefficient. We develop DeNovo, a sequence-based negative sampling and machine learning framework that learns from PPIs of different viruses to predict for a novel one, exploiting the shared host proteins. We tested DeNovo on PPIs from different domains to assess generalization. RESULTS: By solving the challenge of generating less noisy negative interactions, DeNovo achieved accuracy up to 81 and 86% when predicting PPIs of viral proteins that have no and distant sequence similarity to the ones used for training, receptively. This result is comparable to the best achieved in single virus-host and intra-species PPI prediction cases. Thus, we can now predict PPIs for virtually any virus infecting human. DeNovo generalizes well; it achieved near optimal accuracy when tested on bacteria-human interactions. AVAILABILITY AND IMPLEMENTATION: Code, data and additional supplementary materials needed to reproduce this study are available at: https://bioinformatics.cs.vt.edu/~alzahraa/denovo CONTACT: alzahraa@vt.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.
Asunto(s)
Mapeo de Interacción de Proteínas , Proteínas Virales , Virus , Predicción , Humanos , Análisis de Secuencia de ADNRESUMEN
Broadening gene therapy applications requires manufacturable vectors that efficiently transduce target cells in humans and preclinical models. Conventional selections of adeno-associated virus (AAV) capsid libraries are inefficient at searching the vast sequence space for the small fraction of vectors possessing multiple traits essential for clinical translation. Here, we present Fit4Function, a generalizable machine learning (ML) approach for systematically engineering multi-trait AAV capsids. By leveraging a capsid library that uniformly samples the manufacturable sequence space, reproducible screening data are generated to train accurate sequence-to-function models. Combining six models, we designed a multi-trait (liver-targeted, manufacturable) capsid library and validated 88% of library variants on all six predetermined criteria. Furthermore, the models, trained only on mouse in vivo and human in vitro Fit4Function data, accurately predicted AAV capsid variant biodistribution in macaque. Top candidates exhibited production yields comparable to AAV9, efficient murine liver transduction, up to 1000-fold greater human hepatocyte transduction, and increased enrichment relative to AAV9 in a screen for liver transduction in macaques. The Fit4Function strategy ultimately makes it possible to predict cross-species traits of peptide-modified AAV capsids and is a critical step toward assembling an ML atlas that predicts AAV capsid performance across dozens of traits.
Asunto(s)
Proteínas de la Cápside , Cápside , Dependovirus , Vectores Genéticos , Hígado , Dependovirus/genética , Animales , Humanos , Ratones , Vectores Genéticos/genética , Cápside/metabolismo , Proteínas de la Cápside/genética , Proteínas de la Cápside/metabolismo , Hígado/metabolismo , Transducción Genética , Técnicas de Transferencia de Gen , Aprendizaje Automático , Terapia Genética/métodos , Macaca , Hepatocitos/metabolismo , Células HEK293 , Ingeniería Genética/métodosRESUMEN
Developing vehicles that efficiently deliver genes throughout the human central nervous system (CNS) will broaden the range of treatable genetic diseases. We engineered an adeno-associated virus (AAV) capsid, BI-hTFR1, that binds human transferrin receptor (TfR1), a protein expressed on the blood-brain barrier. BI-hTFR1 was actively transported across human brain endothelial cells and, relative to AAV9, provided 40 to 50 times greater reporter expression in the CNS of human TFRC knockin mice. The enhanced tropism was CNS-specific and absent in wild-type mice. When used to deliver GBA1, mutations of which cause Gaucher disease and are linked to Parkinson's disease, BI-hTFR1 substantially increased brain and cerebrospinal fluid glucocerebrosidase activity compared with AAV9. These findings establish BI-hTFR1 as a potential vector for human CNS gene therapy.
Asunto(s)
Antígenos CD , Encéfalo , Cápside , Técnicas de Transferencia de Gen , Vectores Genéticos , Glucosilceramidasa , Receptores de Transferrina , Animales , Humanos , Ratones , Antígenos CD/metabolismo , Antígenos CD/genética , Barrera Hematoencefálica/metabolismo , Encéfalo/metabolismo , Cápside/metabolismo , Proteínas de la Cápside/metabolismo , Proteínas de la Cápside/genética , Dependovirus , Células Endoteliales/metabolismo , Técnicas de Sustitución del Gen , Terapia Genética , Receptores de Transferrina/metabolismo , Receptores de Transferrina/genética , Glucosilceramidasa/genética , Enfermedad de Gaucher/genética , Enfermedad de Gaucher/terapia , Enfermedad de Parkinson/genética , Enfermedad de Parkinson/terapiaRESUMEN
Developing vehicles that efficiently deliver genes throughout the human central nervous system (CNS) will broaden the range of treatable genetic diseases. We engineered an AAV capsid, BI-hTFR1, that binds human Transferrin Receptor (TfR1), a protein expressed on the blood-brain barrier (BBB). BI-hTFR1 was actively transported across a human brain endothelial cell layer and, relative to AAV9, provided 40-50 times greater reporter expression in the CNS of human TFRC knock-in mice. The enhanced tropism was CNS-specific and absent in wild type mice. When used to deliver GBA1, mutations of which cause Gaucher disease and are linked to Parkinson's disease, BI-hTFR1 substantially increased brain and cerebrospinal fluid glucocerebrosidase activity compared to AAV9. These findings establish BI-hTFR1 as a promising vector for human CNS gene therapy.
RESUMEN
Biases in data used to train machine learning (ML) models can inflate their prediction performance and confound our understanding of how and what they learn. Although biases are common in biological data, systematic auditing of ML models to identify and eliminate these biases is not a common practice when applying ML in the life sciences. Here we devise a systematic, principled, and general approach to audit ML models in the life sciences. We use this auditing framework to examine biases in three ML applications of therapeutic interest and identify unrecognized biases that hinder the ML process and result in substantially reduced model performance on new datasets. Ultimately, we show that ML models tend to learn primarily from data biases when there is insufficient signal in the data to learn from. We provide detailed protocols, guidelines, and examples of code to enable tailoring of the auditing framework to other biomedical applications.
Asunto(s)
Minería de Datos , Aprendizaje Automático , Proteínas/metabolismo , Proteoma , Proteómica , Animales , Sesgo , Bases de Datos de Proteínas , Antígenos de Histocompatibilidad/metabolismo , Humanos , Preparaciones Farmacéuticas/química , Preparaciones Farmacéuticas/metabolismo , Unión Proteica , Mapas de Interacción de Proteínas , Proteínas/química , Reproducibilidad de los ResultadosRESUMEN
Planthoppers are the most notorious rice pests, because they transmit various rice viruses in a persistent-propagative manner. Protein-protein interactions (PPIs) between virus and vector are crucial for virus transmission by vector insects. However, the number of known PPIs for pairs of rice viruses and planthoppers is restricted by low throughput research methods. In this study, we applied DeNovo, a virus-host sequence-based PPI predictor, to predict potential PPIs at a genome-wide scale between three planthoppers and five rice viruses. PPIs were identified at two different confidence thresholds, referred to as low and high modes. The number of PPIs for the five planthopper-virus pairs ranged from 506 to 1985 in the low mode and from 1254 to 4286 in the high mode. After eliminating the "one-too-many" redundant interacting information, the PPIs with unique planthopper proteins were reduced to 343-724 in the low mode and 758-1671 in the high mode. Homologous analysis showed that 11 sets and 31 sets of homologous planthopper proteins were shared by all planthopper-virus interactions in the two modes, indicating that they are potential conserved vector factors essential for transmission of rice viruses. Ten PPIs between small brown planthopper and rice stripe virus (RSV) were verified using glutathione-S-transferase (GST)/His-pull down or co-immunoprecipitation assay. Five of the ten PPIs were proven positive, and three of the five SBPH proteins were confirmed to interact with RSV. The predicted PPIs provide new clues for further studies of the complicated relationship between rice viruses and their vector insects.
Asunto(s)
Hemípteros/virología , Interacciones Microbiota-Huesped , Oryza/virología , Virus de Plantas , Animales , Hemípteros/genética , Hemípteros/metabolismo , Inmunoprecipitación/métodos , Proteínas de Insectos/metabolismo , Insectos Vectores/genética , Insectos Vectores/metabolismo , Insectos Vectores/virología , Oryza/metabolismo , Enfermedades de las Plantas/virología , Virus de Plantas/genética , Virus de Plantas/metabolismo , Mapas de Interacción de Proteínas , Tenuivirus/genética , Tenuivirus/metabolismoRESUMEN
With abundance of biological data, computational prediction of gene regulatory networks (GRNs) from gene expression data has become more feasible. Although incorporating other prior knowledge (PK), along with gene expression data, greatly improves prediction accuracy, the overall accuracy is still low. PK in GRN inference can be categorized into noisy and curated. In noisy PK, relations between genes do not necessarily correspond to regulatory relations and are thus considered inaccurate by inference algorithms such as transcription factor binding and protein-protein interactions. In contrast, curated PK is experimentally verified regulatory interactions in pathway databases. An issue in real data is that gene expression can poorly support the curated PK and thus most existing prediction algorithms cannot use these curated PK. Although several algorithms were proposed to incorporate noisy PK, none address curated PK with poor gene expression support. We present PEAK, a system to integrate both curated and noisy PK in GRN inference, especially with poor gene expression support. We introduce a novel method for GRN inference, CurInf, to effectively integrate curated PK, even when the gene expression data poorly support the PK. PEAK also uses the previously proposed method Modified Elastic Net to incorporate noisy PK, and we call it NoisInf. In our experiment, CurInf significantly incorporates curated PK, which was regarded as noise by previous methods. Using 100% curated PK, CurInf improves the area under precision-recall curve accuracy score over NoisInf by 27.3% in synthetic data, 86.5% in Escherichia coli data, and 31.1% in Saccharomyces cerevisiae data. Moreover, even when the noise in PK is 10 times more than true PK, PEAK performs better than inference without any PK. Better integration of curated PK helps biologists benefit from verified experimental data to predict more reliable GRN.