Exploring “dark-matter” protein folds using deep learning

Piscis: A loss estimator of the F1 score enables accurate spot detection in fluorescence microscopy images via deep learning.

This work proposes a hybrid system that combines ResNet (Residual Network - widely recognized for its impact on deep learning, being a milestone in the area of computer vision) with Extremely Random Trees (Extra Trees) to classify chest X-ray images and assist in the detection of diseases. The approach uses the Transfer Learning technique, where ResNet, previously trained, is used to extract relevant characteristics from the images. Then, the Extra Trees algorithm performs the classification based on these characteristics. In the initial stage, using only ResNet combined with a small neural network, we obtained an accuracy of 95.40% in validation and 79.33% in tests. With the implementation of the hybrid system, the results were significantly improved, reaching 96.90% accuracy in validation and 89.98% in tests, representing a significant improvement of approximately 10 percentage points in tests. These results highlight the potential of the hybrid system in applications, demonstrating how the combination of advanced deep learning and machine learning techniques can contribute significantly to improving accuracy.

Retrieving the layout of different cluttered indoor scenes from monocular images is a challenging task, using a smaller deep neural network than the existing proposals. Previous geometric solutions are prone to failure in the presence of cluttered scenes because they depend strongly on hand-engineering features and the expectation of the possibility of the vanishing points’ calculation. With the growth of neural networks, the geometric methods were either replaced or fused within the emerging area of deep learning. The more recent solutions rely on dense neural networks with additional adjustments, either by the calculation of the vanishing points, position based, or by layout ranking. All these methods presented valid solutions to this challenge with the flaw of being computationally demanding. Here, we present a more lightweight solution, running the segmentation on a smaller neural network and introducing a discriminative classifier for the posterior layout ranking and optimization. Our proposed method is evaluated by two standard dataset benchmarks, achieving near state of the art results even with a fraction of the required parameters than the available state of the art methods.

Decoding the role of the arginine dihydrolase pathway in shaping human gut community assembly and health-relevant metabolites.

Classification of triple-negative breast cancers through a Boolean network model of the epithelial-mesenchymal transition.

Evolution in microbial microcosms is highly parallel, regardless of the presence of interacting species

Protein surface chemistry encodes an adaptive tolerance to desiccation.

Learning protein fitness landscapes with deep mutational scanning data from multiple sources.

Susceptibility to metabolic syndrome (MetS) is dependent on genetics, environment, and gene-by-environment interactions, rendering the study of underlying mechanisms challenging. The majority of experiments in model organisms do not incorporate genetic variation and lack specific evaluation criteria for MetS. Here, we derived a continuous metric, the metabolic health score (MHS), based on standard clinical parameters and defined its molecular signatures in the liver and circulation. In human UK Biobank, the MHS associated with MetS status and was predictive of future disease incidence, even in individuals without MetS. Using quantitative trait locus analyses in mice, we found two MHS-associated genetic loci and replicated them in unrelated mouse populations. Through a prioritization scheme in mice and human genetic data, we identified TNKS and MCPH1 as candidates mediating differences in the MHS. Our findings provide insights into the molecular mechanisms sustaining metabolic health across species and uncover likely regulators. A record of this paper's transparent peer review process is included in the supplemental information.

Convolutions are competitive with transformers for protein sequence pretraining

Stimulus-response signaling dynamics characterize macrophage polarization states

Here, we present a method for expressing multiple open reading frames (ORFs) from single transcripts using the leaky scanning model of translation initiation. In this approach termed "stoichiometric expression of mRNA polycistrons by eukaryotic ribosomes" (SEMPER), adjacent ORFs are translated from a single mRNA at tunable ratios determined by their order in the sequence and the strength of their translation initiation sites. We validate this approach by expressing up to three fluorescent proteins from one plasmid in two different cell lines. We then use it to encode a stoichiometrically tuned polycistronic construct encoding gas vesicle acoustic reporter genes that enables efficient formation of the multi-protein complex while minimizing cellular toxicity. We also demonstrate that SEMPER enables polycistronic expression of recombinant monoclonal antibodies from plasmid DNA and of two fluorescent proteins from single mRNAs made through invitro transcription. Finally, we provide a probabilistic model to elucidate the mechanisms underlying SEMPER. A record of this paper's transparent peer review process is included in the supplemental information.

Metabolite concentration changes can have broad consequences on the function and robustness of metabolic networks. Here, we measured the metabolome response of 1,515 CRISPR interference (CRISPRi) E. coli strains targeting all genes in the iML1515 metabolic model. Metabolites that are hardly measurable in wild-type E. coli accumulated in specific CRISPRi strains, indicating that they are normally maintained at low levels. We confirmed metabolite accumulation using liquid chromatography-tandem mass spectrometry (LC-MS/MS) and generated putative reference spectra for 102 metabolites for which no MS2 data had previously been available. We show that minimal metabolite levels are beneficial because they (1) enable substrate level regulation of enzyme activity, (2) prevent competitive inhibition, and (3) suppress side reactions. However, minimal metabolite pools also limit flux through engineered pathways. For example, low levels of farnesyl diphosphate (frdp) constrained a synthetic carotenoid pathway, and we show that the knockdown of octaprenyl diphosphate synthase (IspB) increased frdp levels and carotenoid production. A record of this paper's transparent peer review process is included in the supplemental information.

Entrainment and multi-stability of the p53 oscillator in human cells

Interpretable data integration for single-cell and spatial multi-omics.