Research Briefing in 2023

In 1858, the first standard for microscope objectives was established to encourage interchangeable components. Over the following 150 years, standards have evolved to constrain the size of objectives, which limits the parameters of working distance, field of view and resolution. A new design breaks out of this conventional envelope, offering an ultra-long working distance in air and enabling new neuroscience experiments.

We have developed a framework for the analysis of multi-batch proteome profiling data using isobaric mass tags. Our framework improves quantitative accuracy and increases statistical power by accounting for known sources of variation between batches, thus enabling multiplexed proteome profiling analysis to be performed on large numbers of samples and population cohorts.

Tracking cells is a time-consuming part of biological image analysis, and traditional manual annotation methods are prohibitively laborious for tracking neurons in the deforming and moving Caenorhabditis elegans brain. By leveraging machine learning to develop a ‘targeted augmentation’ method, we substantially reduced the number of labeled images required for tracking.

We developed MAbID, a method for combined genomic profiling of histone modifications and chromatin-binding proteins in single cells, enabling researchers to study the interconnectivity between gene-regulatory mechanisms. We demonstrated MAbID’s implementation in profiling multifactorial changes in chromatin signatures during in vitro neural differentiation and in primary mouse bone marrow tissue.

Although single-cell RNA-sequencing has revolutionized biomedical research, exploring cell states from an extracellular vesicle viewpoint has remained elusive. We present an algorithm, SEVtras, that accurately captures signals from small extracellular vesicles and determines source cell-type secretion activity. SEVtras unlocks an extracellular dimension for single-cell analysis with diagnostic potential.

We developed a machine learning model, RoseTTAFoldNA, that can predict the structures of protein–DNA and protein–RNA complexes. Our model is capable of predicting accurate structures of protein families for which structural information is unknown.

Fluorescent actinometers enable the measurement of light intensity even in the depths of samples and over wide ranges of wavelengths and intensities. We introduce two protocols to quantitatively characterize the spatial distribution of light of various fluorescence imaging systems and to calibrate the illumination of commercially available instruments and light sources.

Single-cell inference of class-switch recombination (sciCSR) is a computational method that analyzes single-cell RNA sequencing data to deduce the temporal trajectory of how B cells develop antibody response.

We developed fatigue-resistant hydrogel optical fibers through the controlled growth of polymeric nanocrystalline domains to enable light delivery to peripheral nerves during locomotion. The hydrogel fibers withstand locomotion strain across more than 30,000 fiber stretch cycles and enable the optogenetic inhibition of pain hypersensitivity in naturally behaving mice.

Here we developed synthetic transactivation domains (TADs) built from human mechanosensitive transcription factors (MTFs). By linking MTF TAD segments together, we engineered compact and potent multipartite transcriptional activation modules. We then harnessed these modules to create a CRISPR activation system, which we termed the dCas9 recruited enhanced activation module (CRISPR-DREAM).

CryoREAD automatically builds DNA–RNA atomic structure from cryo-EM maps. Backbone accuracy is typically >85% and the method is applicable for maps with RNA-only, DNA-only and DNA–RNA–protein complex structures. CryoREAD uses deep learning to identify structure information and subsequently construct the 3D structure of nucleic acids.

We conducted a comprehensive long-read RNA sequencing (RNA-seq) benchmarking experiment by combining spike-ins and in silico mixtures to establish a ground-truth dataset. We used long- and short-read RNA-seq technology to deeply sequence samples and compared the performance of a range of analysis tools on these data.

We developed CellOT, a tool that integrates optimal transport with input convex neural networks to predict molecular responses of individual cells to various perturbations. By learning a map between the unpaired distributions of unperturbed and perturbed cells, CellOT outperforms current methods and generalizes the inference of treatment outcomes in unobserved cell types and patients.

Modern high-throughput metagenomics is producing hundreds of thousands of metagenome-assembled genomes (MAGs), which is overwhelming traditional sequence-similarity search methods. We present a computational method, skani, that efficiently compares MAGs on a terabyte scale while being robust to the inherent noise in MAGs, enabling larger and more accurate analyses.

To capture expansive, seamless fields of view from frozen hydrated specimens by cryo-electron tomography, we developed methods for the collection and processing of montage data. This approach enables rapid acquisition of contiguous regions of specimens using a montaged tilt series collection scheme.

We introduce GelMap, a flexible workflow for reporting deformations and anisotropy in expansion microscopy. By intrinsically calibrating the expansion hydrogel using a fluorescent grid that scales with expansion and deforms with anisotropy, GelMap enables the reliable quantification of expansion factors and correction of deformations.

Leveraging nanopore long-read sequencing, scNanoHi-C identifies multiway interactions between enhancers and their target promoters within a single cell. Compared with short-read-based single-cell Hi-C or population-based multiway sequencing methods, scNanoHi-C offers new opportunities to investigate the heterogeneities of single-cell gene regulation networks mediated by high-order 3D chromatin structures.

The conversion of biological molecules into digital signals through sequencing is a complex process that often generates substantial systematic background noise. This noise can obscure important biological insights. However, by precisely identifying and removing this noise, we can bring the true signal into focus and eliminate misleading results from downstream analyses.

Single-cell perturbation screens are routinely conducted to study the effects of different perturbations on cellular state, yet such studies are easily confounded by nuisance sources of variation shared with control cells. We present a deep learning method that isolates perturbation-specific sources of variation, enabling a better understanding of the perturbation’s effects.

CheckM2 is a tool that applies machine learning to evaluate the quality of genomes from metagenomic data. CheckM2 is faster and more accurate than existing methods, and it outperforms them when applied to novel lineages and lineages with reduced genome sizes, such as Patescibacteria and the DPANN superphylum.