Discovery of antimicrobial peptides in the global microbiome with machine learning.

Novel antibiotics are urgently needed to combat the antibiotic-resistance crisis. We present a machine-learning-based approach to predict antimicrobial peptides (AMPs) within the global microbiome and leverage a vast dataset of 63,410 metagenomes and 87,920 prokaryotic genomes from environmental and host-associated habitats to create the AMPSphere, a comprehensive catalog comprising 863,498 non-redundant peptides, few of which match existing databases. AMPSphere provides insights into the evolutionary origins of peptides, including by duplication or gene truncation of longer sequences, and we observed that AMP production varies by habitat. To validate our predictions, we synthesized and tested 100 AMPs against clinically relevant drug-resistant pathogens and human gut commensals both in vitro and in vivo. A total of 79 peptides were active, with 63 targeting pathogens. These active AMPs exhibited antibacterial activity by disrupting bacterial membranes. In conclusion, our approach identified nearly one million prokaryotic AMP sequences, an open-access resource for antibiotic discovery.

Towards the biogeography of prokaryotic genes.

Microbial genes encode the majority of the functional repertoire of life on earth. However, despite increasing efforts in metagenomic sequencing of various habitats1-3, little is known about the distribution of genes across the global biosphere, with implications for human and planetary health. Here we constructed a non-redundant gene catalogue of 303 million species-level genes (clustered at 95% nucleotide identity) from 13,174 publicly available metagenomes across 14 major habitats and use it to show that most genes are specific to a single habitat. The small fraction of genes found in multiple habitats is enriched in antibiotic-resistance genes and markers for mobile genetic elements. By further clustering these species-level genes into 32 million protein families, we observed that a small fraction of these families contain the majority of the genes (0.6% of families account for 50% of the genes). The majority of species-level genes and protein families are rare. Furthermore, species-level genes, and in particular the rare ones, show low rates of positive (adaptive) selection, supporting a model in which most genetic variability observed within each protein family is neutral or nearly neutral.

STAB2: an updated spatio-temporal cell atlas of the human and mouse brain.

The brain is constituted of heterogeneous types of neuronal and non-neuronal cells, which are organized into distinct anatomical regions, and show precise regulation of gene expression during development, aging and function. In the current database release, STAB2 provides a systematic cellular map of the human and mouse brain by integrating recently published large-scale single-cell and single-nucleus RNA-sequencing datasets from diverse regions and across lifespan. We applied a hierarchical strategy of unsupervised clustering on the integrated single-cell transcriptomic datasets to precisely annotate the cell types and subtypes in the human and mouse brain. Currently, STAB2 includes 71 and 61 different cell subtypes defined in the human and mouse brain, respectively. It covers 63 subregions and 15 developmental stages of human brain, and 38 subregions and 30 developmental stages of mouse brain, generating a comprehensive atlas for exploring spatiotemporal transcriptomic dynamics in the mammalian brain. We also augmented web interfaces for querying and visualizing the gene expression in specific cell types. STAB2 is freely available at https://mai.fudan.edu.cn/stab2.

An augmented Mendelian randomization approach provides causality of brain imaging features on complex traits in a single biobank-scale dataset.

Mendelian randomization (MR) is an effective approach for revealing causal risk factors that underpin complex traits and diseases. While MR has been more widely applied under two-sample settings, it is more promising to be used in one single large cohort given the rise of biobank-scale datasets that simultaneously contain genotype data, brain imaging data, and matched complex traits from the same individual. However, most existing multivariable MR methods have been developed for two-sample setting or a small number of exposures. In this study, we introduce a one-sample multivariable MR method based on partial least squares and Lasso regression (MR-PL). MR-PL is capable of considering the correlation among exposures (e.g., brain imaging features) when the number of exposures is extremely upscaled, while also correcting for winner's curse bias. We performed extensive and systematic simulations, and demonstrated the robustness and reliability of our method. Comprehensive simulations confirmed that MR-PL can generate more precise causal estimates with lower false positive rates than alternative approaches. Finally, we applied MR-PL to the datasets from UK Biobank to reveal the causal effects of 36 white matter tracts on 180 complex traits, and showed putative white matter tracts that are implicated in smoking, blood vascular function-related traits, and eating behaviors.

Deciphering the genetic architecture of human brain structure and function: a brief survey on recent advances of neuroimaging genomics.

Brain imaging genomics is an emerging interdisciplinary field, where integrated analysis of multimodal medical image-derived phenotypes (IDPs) and multi-omics data, bridging the gap between macroscopic brain phenotypes and their cellular and molecular characteristics. This approach aims to better interpret the genetic architecture and molecular mechanisms associated with brain structure, function and clinical outcomes. More recently, the availability of large-scale imaging and multi-omics datasets from the human brain has afforded the opportunity to the discovering of common genetic variants contributing to the structural and functional IDPs of the human brain. By integrative analyses with functional multi-omics data from the human brain, a set of critical genes, functional genomic regions and neuronal cell types have been identified as significantly associated with brain IDPs. Here, we review the recent advances in the methods and applications of multi-omics integration in brain imaging analysis. We highlight the importance of functional genomic datasets in understanding the biological functions of the identified genes and cell types that are associated with brain IDPs. Moreover, we summarize well-known neuroimaging genetics datasets and discuss challenges and future directions in this field.

A survey on computational strategies for genome-resolved gut metagenomics.

Recovering high-quality metagenome-assembled genomes (HQ-MAGs) is critical for exploring microbial compositions and microbe-phenotype associations. However, multiple sequencing platforms and computational tools for this purpose may confuse researchers and thus call for extensive evaluation. Here, we systematically evaluated a total of 40 combinations of popular computational tools and sequencing platforms (i.e. strategies), involving eight assemblers, eight metagenomic binners and four sequencing technologies, including short-, long-read and metaHiC sequencing. We identified the best tools for the individual tasks (e.g. the assembly and binning) and combinations (e.g. generating more HQ-MAGs) depending on the availability of the sequencing data. We found that the combination of the hybrid assemblies and metaHiC-based binning performed best, followed by the hybrid and long-read assemblies. More importantly, both long-read and metaHiC sequencings link more mobile elements and antibiotic resistance genes to bacterial hosts and improve the quality of public human gut reference genomes with 32% (34/105) HQ-MAGs that were either of better quality than those in the Unified Human Gastrointestinal Genome catalog version 2 or novel.

NanoDeep: a deep learning framework for nanopore adaptive sampling on microbial sequencing.

Nanopore sequencers can enrich or deplete the targeted DNA molecules in a library by reversing the voltage across individual nanopores. However, it requires substantial computational resources to achieve rapid operations in parallel at read-time sequencing. We present a deep learning framework, NanoDeep, to overcome these limitations by incorporating convolutional neural network and squeeze and excitation. We first showed that the raw squiggle derived from native DNA sequences determines the origin of microbial and human genomes. Then, we demonstrated that NanoDeep successfully classified bacterial reads from the pooled library with human sequence and showed enrichment for bacterial sequence compared with routine nanopore sequencing setting. Further, we showed that NanoDeep improves the sequencing efficiency and preserves the fidelity of bacterial genomes in the mock sample. In addition, NanoDeep performs well in the enrichment of metagenome sequences of gut samples, showing its potential applications in the enrichment of unknown microbiota. Our toolkit is available at https://github.com/lysovosyl/NanoDeep.

Engineering of cytosine base editors with DNA damage minimization and editing scope diversification.

Cytosine base editors (CBEs), which enable precise C-to-T substitutions, have been restricted by potential safety risks, including DNA off-target edits, RNA off-target edits and additional genotoxicity such as DNA damages induced by double-strand breaks (DSBs). Though DNA and RNA off-target edits have been ameliorated via various strategies, evaluation and minimization of DSB-associated DNA damage risks for most CBEs remain to be resolved. Here we demonstrate that YE1, an engineered CBE variant with minimized DNA and RNA off-target edits, could induce prominent DSB-associated DNA damage risks, manifested as γH2AX accumulation in human cells. We then perform deaminase engineering for two deaminases lamprey LjCDA1 and human APOBEC3A, and generate divergent CBE variants with eliminated DSB-associated DNA damage risks, in addition to minimized DNA/RNA off-target edits. Furthermore, the editing scopes and sequence preferences of APOBEC3A-derived CBEs could be further diversified by internal fusion strategy. Taken together, this study provides updated evaluation platform for DSB-associated DNA damage risks of CBEs and further generates a series of safer toolkits with diversified editing signatures to expand their applications.

A Diffusive Artificial Synapse Based on Charged Metal Nanoparticles.

We show that a diffusive memristor with analogue switching characteristics can be achieved in a layer of gold nanoparticles (AuNPs) functionalized with charged self-assembled monolayers (deprotonated 11-mercaptoundecanoic acid). The nanoparticle core and the anchored stationary charges are jammed within the layer while the mobile counterions [N(CH3)4+] can respond to the electric field and spontaneously diffuse back to the initial positions upon removal of the field. This metal nanoparticle device is set-step free, energy consumption efficient, mechanically flexible, and analogous to bio-Ca2+ dynamics and has tunable conductance modulation capabilities at the counterion concentrations. The gradual resistive switching behavior enables us to implement several important synaptic functions such as potentiation/depression, spike voltage-dependent plasticity, spike duration-dependent plasticity, spike frequency-dependent plasticity, and paired-pulse facilitation. Finally, on the basis of the paired-pulse facilitation characteristics, the metal nanoparticle diffusive artificial synapse is used for edge extraction with exhibits excellent performance.

Probing Oxidation Mechanisms in Plasmonic Catalysis: Unraveling the Role of Reactive Oxygen Species.

Plasmon-induced oxidation has conventionally been attributed to the transfer of plasmonic hot holes. However, this theoretical framework encounters challenges in elucidating the latest experimental findings, such as enhanced catalytic efficiency under uncoupled irradiation conditions and superior oxidizability of silver nanoparticles. Herein, we employ liquid surface-enhanced Raman spectroscopy (SERS) as a real-time and in situ tool to explore the oxidation mechanisms in plasmonic catalysis, taking the decarboxylation of p-mercaptobenzoic acid (PMBA) as a case study. Our findings suggest that the plasmon-induced oxidation is driven by reactive oxygen species (ROS) rather than hot holes, holding true for both the Au and Ag nanoparticles. Subsequent investigations suggest that plasmon-induced ROS may arise from hot carriers or energy transfer mechanisms, exhibiting selectivity under different experimental conditions. The observations were substantiated by investigating the cleavage of the carbon-boron bonds. Furthermore, the underlying mechanisms were clarified by energy level theories, advancing our understanding of plasmonic catalysis.

Medication non-adherence and self-inflicted violence behaviors among 185,800 patients with schizophrenia in the community: a 12-year cohort study.

BACKGROUND: Despite the importance of medication adherence in treatment effectiveness, little is known about the association between medication non-adherence and self-inflicted violence behaviors. We aimed to assess whether medication non-adherence increased the risk of self-inflicted violence behaviors among schizophrenics in communities (hypothesis 1) and whether the dose-response relationship existed (hypothesis 2). METHODS: This 12-year cohort study in western China recruited 292,667 community-dwelling schizophrenics. The proportion of regular medication (PRM) was calculated by dividing the time of "regular adherence" by the total time of antipsychotic treatment during follow-up period as an indicator of medication adherence. For hypothesis 1, medication adherence was designated as a binary variable with a threshold of 0.8 (PRM); for hypothesis 2, medication adherence was specified as five-category and continuous variables, respectively. Inverse probability weighting and mixed effects Cox proportional hazards models were conducted for confounders control and survival analyses. RESULTS: One hundred eighty-five thousand eight hundred participants were eligible for the final analyses, with a mean age of 47.49 years (SD 14.55 years), of whom 53.6% were female. For hypothesis 1, the medication non-adherence group (PRM < 0.8) had a lower risk of suicide (HR, 0.527, 95% CI, 0.447-0.620), an increased risk of NSSI (HR, 1.229, 95% CI, 1.088-1.388), and non-significant risk of attempted suicide compared with adherence group (PRM ≥ 0.8). For hypothesis 2, the lowest medication adherence (PRM < 0.2) was associated with increased risks of suicide attempt (HR, 1.614, 95% CI, 1.412-1.845), NSSI (HR, 1.873, 95% CI, 1.649-2.126), and a decreased risk of suicide (HR, 0.593, 95% CI, 0.490-0.719). The other non-adherence groups had lower risks for all three self-inflicted violence behaviors. The associations between medication adherence in continuous-variable and three outcomes were consistent with the categorical medication adherence results. CONCLUSIONS: Almost no medication taken as prescribed was associated with an increased risk of suicide attempt and NSSI. However, medication adherence did not appear to prevent completed suicide. Besides, patients with moderate adherence had a lower incidence of suicide attempt and NSSI. These findings highlight the need for a more detailed portrayal of medication adherence and the need to be vigilant for suicide intent in schizophrenics with good medication adherence who may be overlooked previously.

Hybrid Cellular Nanovesicles Block PD-L1 Signal and Repolarize M2 Macrophages for Cancer Immunotherapy.

The PD1/PD-L1 immune checkpoint blocking is a promising therapy, while immunosuppressive tumor microenvironment (TME) and poor tumor penetration of therapeutic antibodies limit its efficacy. Repolarization of tumor-associated macrophages (TAMs) offers a potential method to ameliorate immunosuppression of TME and further boost T cell antitumor immunity. Herein, hybrid cell membrane biomimetic nanovesicles (hNVs) are developed by fusing M1 macrophage-derived nanovesicles (M1-NVs) and PD1-overexpressed tumor cell-derived nanovesicles (PD1-NVs) to improve cancer immunotherapy. The M1-NVs promote the transformation of M2-like TAMs to M1-like phenotype and further increase the release of pro-inflammatory cytokines, resulting in improved immunosuppressive TME. Concurrently, the PD1-NVs block PD1/PD-L1 pathway, which boosts cancer immunotherapy when combined with M1-NVs. In a breast cancer mouse model, the hNVs efficiently accumulate at the tumor site after intravenous injection and significantly inhibit the tumor growth. Mechanically, the M1 macrophages and CD8+ T lymphocytes in TME increase by twofold after the treatment, indicating effective immune activation. These results suggest the hNVs as a promising strategy to integrate TME improvement with PD1/PD-L1 blockade for cancer immunotherapy.

Injectable Oxygen-Carrying Microsphere Hydrogel for Dynamic Regulation of Redox Microenvironment of Wounds.

The delayed healing of infected wounds can be attributed to the increased production of reactive oxygen species (ROS) and consequent damages to vascellum and tissue, resulting in a hypoxic wound environment that further exacerbates inflammation. Current clinical treatments including hyperbaric oxygen therapy and antibiotic treatment fail to provide sustained oxygenation and drug-free resistance to infection. To propose a dynamic oxygen regulation strategy, this study develops a composite hydrogel with ROS-scavenging system and oxygen-releasing microspheres in the wound dressing. The hydrogel itself reduces cellular damage by removing ROS derived from immune cells. Simultaneously, the sustained release of oxygen from microspheres improves cell survival and migration in hypoxic environments, promoting angiogenesis and collagen regeneration. The combination of ROS scavenging and oxygenation enables the wound dressing to achieve drug-free anti-infection through activating immune modulation, inhibiting the secretion of pro-inflammatory cytokines interleukin-6, and promoting tissue regeneration in both acute and infected wounds of rat skins. Thus, the composite hydrogel dressing proposed in this work shows great potential for dynamic redox regulation of infected wounds and accelerates wound healing without drugs.

MorbidGCN: prediction of multimorbidity with a graph convolutional network based on integration of population phenotypes and disease network.

Exploring multimorbidity relationships among diseases is of great importance for understanding their shared mechanisms, precise diagnosis and treatment. However, the landscape of multimorbidities is still far from complete due to the complex nature of multimorbidity. Although various types of biological data, such as biomolecules and clinical symptoms, have been used to identify multimorbidities, the population phenotype information (e.g. physical activity and diet) remains less explored for multimorbidity. Here, we present a graph convolutional network (GCN) model, named MorbidGCN, for multimorbidity prediction by integrating population phenotypes and disease network. Specifically, MorbidGCN treats the multimorbidity prediction as a missing link prediction problem in the disease network, where a novel feature selection method is embedded to select important phenotypes. Benchmarking results on two large-scale multimorbidity data sets, i.e. the UK Biobank (UKB) and Human Disease Network (HuDiNe) data sets, demonstrate that MorbidGCN outperforms other competitive methods. With MorbidGCN, 9742 and 14 010 novel multimorbidities are identified in the UKB and HuDiNe data sets, respectively. Moreover, we notice that the selected phenotypes that are generally differentially distributed between multimorbidity patients and single-disease patients can help interpret multimorbidities and show potential for prognosis of multimorbidities.

An integrated brain-specific network identifies genes associated with neuropathologic and clinical traits of Alzheimer's disease.

Alzheimer's disease (AD) has a strong genetic predisposition. However, its risk genes remain incompletely identified. We developed an Alzheimer's brain gene network-based approach to predict AD-associated genes by leveraging the functional pattern of known AD-associated genes. Our constructed network outperformed existing networks in predicting AD genes. We then systematically validated the predictions using independent genetic, transcriptomic, proteomic data, neuropathological and clinical data. First, top-ranked genes were enriched in AD-associated pathways. Second, using external gene expression data from the Mount Sinai Brain Bank study, we found that the top-ranked genes were significantly associated with neuropathological and clinical traits, including the Consortium to Establish a Registry for Alzheimer's Disease score, Braak stage score and clinical dementia rating. The analysis of Alzheimer's brain single-cell RNA-seq data revealed cell-type-specific association of predicted genes with early pathology of AD. Third, by interrogating proteomic data in the Religious Orders Study and Memory and Aging Project and Baltimore Longitudinal Study of Aging studies, we observed a significant association of protein expression level with cognitive function and AD clinical severity. The network, method and predictions could become a valuable resource to advance the identification of risk genes for AD.

SemiBin2: self-supervised contrastive learning leads to better MAGs for short- and long-read sequencing.

MOTIVATION: Metagenomic binning methods to reconstruct metagenome-assembled genomes (MAGs) from environmental samples have been widely used in large-scale metagenomic studies. The recently proposed semi-supervised binning method, SemiBin, achieved state-of-the-art binning results in several environments. However, this required annotating contigs, a computationally costly and potentially biased process. RESULTS: We propose SemiBin2, which uses self-supervised learning to learn feature embeddings from the contigs. In simulated and real datasets, we show that self-supervised learning achieves better results than the semi-supervised learning used in SemiBin1 and that SemiBin2 outperforms other state-of-the-art binners. Compared to SemiBin1, SemiBin2 can reconstruct 8.3-21.5% more high-quality bins and requires only 25% of the running time and 11% of peak memory usage in real short-read sequencing samples. To extend SemiBin2 to long-read data, we also propose ensemble-based DBSCAN clustering algorithm, resulting in 13.1-26.3% more high-quality genomes than the second best binner for long-read data. AVAILABILITY AND IMPLEMENTATION: SemiBin2 is available as open source software at https://github.com/BigDataBiology/SemiBin/ and the analysis scripts used in the study can be found at https://github.com/BigDataBiology/SemiBin2_benchmark.

CD19/CD20 dual-targeted chimeric antigen receptor-engineered natural killer cells exhibit improved cytotoxicity against acute lymphoblastic leukemia.

BACKGROUND: Chimeric antigen receptor natural killer (CAR-NK) cells represent a promising advancement in CAR cell therapy, addressing limitations observed in CAR-T cell therapy. However, our prior study revealed challenges in CAR-NK cells targeting CD19 antigens, as they failed to eliminate CD19+ Raji cells in NSG tumor-bearing mice, noting down-regulation or loss of CD19 antigen expression in some Raji cells. In response, this study aims to enhance CD19 CAR-NK cell efficacy and mitigate the risk of tumor recurrence due to target antigen escape by developing CD19 and CD20 (CD19/CD20) dual-targeted CAR-NK cells. METHODS: Initially, mRNA encoding anti-CD19 CARs (FMC63 scFv-CD8α-4-1BB-CD3ζ) and anti-CD20 CARs (LEU16 scFv-CD8α-4-1BB-CD3ζ) was constructed via in vitro transcription. Subsequently, CD19/CD20 dual-targeted CAR-NK cells were generated through simultaneous electrotransfection of CD19/CD20 CAR mRNA into umbilical cord blood-derived NK cells (UCB-NK). RESULTS: Following co-electroporation, the percentage of dual-CAR expression on NK cells was 86.4% ± 1.83%, as determined by flow cytometry. CAR expression was detectable at 8 h post-electric transfer, peaked at 24 h, and remained detectable at 96 h. CD19/CD20 dual-targeted CAR-NK cells exhibited increased specific cytotoxicity against acute lymphoblastic leukemia (ALL) cell lines (BALL-1: CD19+CD20+, REH: CD19+CD20-, Jurkat: CD19-CD20-) compared to UCB-NK, CD19 CAR-NK, and CD20 CAR-NK cells. Moreover, CD19/CD20 dual-targeted CAR-NK cells released elevated levels of perforin, IFN-γ, and IL-15. Multiple activation markers such as CD69 and cytotoxic substances were highly expressed. CONCLUSIONS: The creation of CD19/CD20 dual-targeted CAR-NK cells addressed the risk of tumor escape due to antigen heterogeneity in ALL, offering efficient and safe 'off-the-shelf' cell products. These cells demonstrate efficacy in targeting CD20 and/or CD19 antigens in ALL, laying an experimental foundation for their application in ALL treatment.

Effective focal spot measurement method for X-ray source based on the dynamic translation of a light barrier.

The accuracy of measuring the effective focal spot of the X-ray source directly affects the spatial resolution of computed tomography (CT) reconstructed images. This study proposes what we believe to be a novel approach to measure the effective focal spot based on the dynamic translation of light barrier using an accessible measuring device. This method discretizes the effective focal spot of the X-ray source into multiple subfocal spots with varying intensities and establishes a nonlinear model between the effective focal spot and measurement data. Measurement data are obtained by moving the light barrier to different positions using the electric displacement stage. The shape, size, and intensity distribution of the effective focal spot are determined by calculating the normalized weighting coefficients for each subfocal spot from measurement data. The measurement device is simple and easy to operate. Additionally, the obtained effective focal spot exhibits high accuracy, and a higher spatial resolution can be realized by reconstructing the CT images using the measured focal spot information. Numerical and real experiments validate the proposed method.

Quantitative real-time phase microscopy for extended depth-of-field imaging based on the 3D single-shot differential phase contrast (ssDPC) imaging method.

Optical diffraction tomography (ODT) is a promising label-free imaging method capable of quantitatively measuring the three-dimensional (3D) refractive index distribution of transparent samples. In recent years, partially coherent ODT (PC-ODT) has attracted increasing attention due to its system simplicity and absence of laser speckle noise. Quantitative phase imaging (QPI) technologies represented by Fourier ptychographic microscopy (FPM), differential phase contrast (DPC) imaging and intensity diffraction tomography (IDT) need to collect several or hundreds of intensity images, which usually introduce motion artifacts when shooting fast-moving targets, leading to a decrease in image quality. Hence, a quantitative real-time phase microscopy (qRPM) for extended depth of field (DOF) imaging based on 3D single-shot differential phase contrast (ssDPC) imaging method is proposed in this research study. qRPM incorporates a microlens array (MLA) to simultaneously collect spatial information and angular information. In subsequent optical information processing, a deconvolution method is used to obtain intensity stacks under different illumination angles in a raw light field image. Importing the obtained intensity stack into the 3D DPC imaging model is able to finally obtain the 3D refractive index distribution. The captured four-dimensional light field information enables the reconstruction of 3D information in a single snapshot and extending the DOF of qRPM. The imaging capability of the proposed qRPM system is experimental verified on different samples, achieve single-exposure 3D label-free imaging with an extended DOF for 160 µm which is nearly 30 times higher than the traditional microscope system.

High performance few-mode fiber-based light field direction sensing system using deep convolutional neural network: fiber speckle demodulation network (FSDNET).

Precisely sensing the light field direction information plays the essential role in the fields of three-dimensional (3D) imaging, light field sensing, target positioning and tracking, remote sensing, etc. It is thrilling to find that the optical fiber can be used as a sensing component due to its high sensitivity, compact size, and strong resistance to electromagnetic interference. According to the core principle that the few-mode fiber output speckle pattern is sensitive to the change of incident light field direction, the variation characteristics is further investigated in this research study. Based on the simulation and analysis of the fiber transmission characteristics, the output speckle corresponding to the incident light field with the direction in the range of ±6° horizontally and vertically are calculated. Furthermore, a deep convolutional neural network (CNN): fiber speckle demodulation network (FSDNET) is proposed and constructed to establish what we believe to be a novel way to reveal and identify the mapping relationship between the light field direction and the output speckle. The theoretical simulation shows that the mean absolute error (MAE) between the perceived light field directions and the true directions is 0.01°. Then, a light field direction sensing system based on the few-mode fiber is developed. Regarding to the performance of the sensing system, the MAE of the FSDNET for the light field directions that have appeared in the training set is 0.0389°, and for testing set of the unknown directions that have not appeared in the training set, the MAE is 0.0570°. Therefore, the simulation and experimental results prove that high performance sensing of light field direction can be achieved by the proposed few-mode fiber sensing system and the FSDNET.