Deep Learning Model for Prediction of Bronchopulmonary Dysplasia in Preterm Infants Using Chest Radiographs.

Bronchopulmonary dysplasia (BPD) is common in preterm infants and may result in pulmonary vascular disease, compromising lung function. This study aimed to employ artificial intelligence (AI) techniques to help physicians accurately diagnose BPD in preterm infants in a timely and efficient manner. This retrospective study involves two datasets: a lung region segmentation dataset comprising 1491 chest radiographs of infants, and a BPD prediction dataset comprising 1021 chest radiographs of preterm infants. Transfer learning of a pre-trained machine learning model was employed for lung region segmentation and image fusion for BPD prediction to enhance the performance of the AI model. The lung segmentation model uses transfer learning to achieve a dice score of 0.960 for preterm infants with ≤ 168 h postnatal age. The BPD prediction model exhibited superior diagnostic performance compared to that of experts and demonstrated consistent performance for chest radiographs obtained at ≤ 24 h postnatal age, and those obtained at 25 to 168 h postnatal age. This study is the first to use deep learning on preterm chest radiographs for lung segmentation to develop a BPD prediction model with an early detection time of less than 24 h. Additionally, this study compared the model's performance according to both NICHD and Jensen criteria for BPD. Results demonstrate that the AI model surpasses the diagnostic accuracy of experts in predicting lung development in preterm infants.

A novel deep learning-based algorithm combining histopathological features with tissue areas to predict colorectal cancer survival from whole-slide images.

BACKGROUND: Many methodologies for selecting histopathological images, such as sample image patches or segment histology from regions of interest (ROIs) or whole-slide images (WSIs), have been utilized to develop survival models. With gigapixel WSIs exhibiting diverse histological appearances, obtaining clinically prognostic and explainable features remains challenging. Therefore, we propose a novel deep learning-based algorithm combining tissue areas with histopathological features to predict cancer survival. METHODS: The Cancer Genome Atlas Colon Adenocarcinoma (TCGA-COAD) dataset was used in this investigation. A deep convolutional survival model (DeepConvSurv) extracted histopathological information from the image patches of nine different tissue types, including tumors, lymphocytes, stroma, and mucus. The tissue map of the WSIs was segmented using image processing techniques that involved localizing and quantifying the tissue region. Six survival models with the concordance index (C-index) were used as the evaluation metrics. RESULTS: We extracted 128 histopathological features from four histological types and five tissue area features from WSIs to predict colorectal cancer survival. Our method performed better in six distinct survival models than the Whole Slide Histopathological Images Survival Analysis framework (WSISA), which adaptively sampled patches using K-means from WSIs. The best performance using histopathological features was 0.679 using LASSO-Cox. Compared to histopathological features alone, tissue area features increased the C-index by 2.5%. Based on histopathological features and tissue area features, our approach achieved performance of 0.704 with RIDGE-Cox. CONCLUSIONS: A deep learning-based algorithm combining histopathological features with tissue area proved clinically relevant and effective for predicting cancer survival.

Slice-Fusion: Reducing False Positives in Liver Tumor Detection for Mask R-CNN.

Automatic liver tumor detection from computed tomography (CT) makes clinical examinations more accurate. However, deep learning-based detection algorithms are characterized by high sensitivity and low precision, which hinders diagnosis given that false-positive tumors must first be identified and excluded. These false positives arise because detection models incorrectly identify partial volume artifacts as lesions, which in turn stems from the inability to learn the perihepatic structure from a global perspective. To overcome this limitation, we propose a novel slice-fusion method in which mining the global structural relationship between the tissues in the target CT slices and fusing the features of adjacent slices according to the importance of the tissues. Furthermore, we design a new network based on our slice-fusion method and Mask R-CNN detection model, called Pinpoint-Net. We evaluated proposed model on the Liver Tumor Segmentation Challenge (LiTS) dataset and our liver metastases dataset. Experiments demonstrated that our slice-fusion method not only enhance tumor detection ability via reducing the number of false-positive tumors smaller than 10mm, but also improve segmentation performance. Without bells and whistles, a single Pinpoint-Net showed outstanding performance in liver tumor detection and segmentation on LiTS test dataset compared with other state-of-the-art models.

Detecting Hydronephrosis Through Ultrasound Images Using State-of-the-Art Deep Learning Models.

The goal of this study was to assess the feasibility of three models for detecting hydronephrosis through ultrasound images using state-of-the-art deep learning algorithms. The diagnosis of hydronephrosis is challenging because of varying and non-specific presentations. With the characteristics of ready accessibility, no radiation exposure and repeated assessments, point-of-care ultrasound becomes a complementary diagnostic tool for hydronephrosis; however, inter-observer variability still exists after time-consuming training. Artificial intelligence has the potential to overcome the human limitations. A total of 3462 ultrasound frames for 97 patients with hydronephrosis confirmed by the expert nephrologists were included. One thousand six hundred twenty-eight ultrasound frames were also extracted from the 265 controls who had normal renal ultrasonography. We built three deep learning models based on U-Net, Res-UNet and UNet++ and compared their performance. We applied pre-processing techniques including wiping the background to lessen interference by YOLOv4 and standardizing image sizes. Also, post-processing techniques such as adding filter for filtering the small effusion areas were used. The Res-UNet algorithm had the best performance with an accuracy of 94.6% for moderate/severe hydronephrosis with substantial recall rate, specificity, precision, F1 measure and intersection over union. The Res-UNet algorithm has the best performance in detection of moderate/severe hydronephrosis. It would decrease variability among sonographers and improve efficiency under clinical conditions.

Pathogenicity Prediction of Single Amino Acid Variants With Machine Learning Model Based on Protein Structural Energies.

The most popular tools for predicting pathogenicity of single amino acid variants (SAVs) were developed based on sequence-based techniques. SAVs may change protein structure and function. In the context of van der Waals force and disulfide bridge calculations, no method directly predicts the impact of mutations on the energies of the protein structure. Here, we combined machine learning methods and energy scores of protein structures calculated by Rosetta Energy Function 2015 to predict SAV pathogenicity. The accuracy level of our model (0.76) is higher than that of six prediction tools. Further analyses revealed that the differential reference energies, attractive energies, and solvation of polar atoms between wildtype and mutant side-chains played essential roles in distinguishing benign from pathogenic variants. These features indicated the physicochemical properties of amino acids, which were observed in 3D structures instead of sequences. We added 16 features to Rhapsody (the prediction tool we used for our data set) and consequently improved its performance. The results indicated that these energy scores were more appropriate and more detailed representations of the pathogenicity of SAVs.

Novel Algorithm for Improved Protein Classification Using Graph Similarity.

Considerable sequence data are produced in genome annotation projects that relate to molecular levels, structural similarities, and molecular and biological functions. In structural genomics, the most essential task involves resolving protein structures efficiently with hardware or software, understanding these structures, and assigning their biological functions. Understanding the characteristics and functions of proteins enables the exploration of the molecular mechanisms of life. In this paper, we examine the problems of protein classification. Because they perform similar biological functions, proteins in the same family usually share similar structural characteristics. We employed this premise in designing a classification algorithm. In this algorithm, auxiliary graphs are used to represent proteins, with every amino acid in a protein to a vertex in a graph. Moreover, the links between amino acids correspond to the edges between the vertices. The proposed algorithm classifies proteins according to the similarities in their graphical structures. The proposed algorithm is efficient and accurate in distinguishing proteins from different families and outperformed related algorithms experimentally.

Sensitive and Specific Cadmium Biosensor Developed by Reconfiguring Metal Transport and Leveraging Natural Gene Repositories.

Whole-cell biosensors are useful for monitoring heavy metal toxicity in public health and ecosystems, but their development has been hindered by intrinsic trade-offs between sensitivity and specificity. Here, we demonstrated an effective engineering solution by building a sensitive, specific, and high-response biosensor for carcinogenic cadmium ions. We genetically programmed the metal transport system of Escherichia coli to enrich intracellular cadmium ions and deprive interfering metal species. We then selected 16 cadmium-sensing transcription factors from the GenBank database and tested their reactivity to 14 metal ions in the engineered E. coli using the expression of the green fluorescent protein as the readout. The resulting cadmium biosensor was highly specific and showed a detection limit of 3 nM, a linear increase in fluorescent intensities from 0 to 200 nM, and a maximal 777-fold signal change. Using this whole-cell biosensor, a smartphone, and low-tech equipment, we developed a simple assay capable of measuring cadmium ions at the same concentration range in irrigation water and human urine. This method is user-friendly and cost-effective, making it affordable to screen large amounts of samples for cadmium toxicity in agriculture and medicine. Moreover, our work highlights natural gene repositories as a treasure chest for bioengineering.

A Novel Branch-and-Bound Algorithm for the Protein Folding Problem in the 3D HP Model.

The protein folding problem (PFP) is an important issue in bioinformatics and biochemical physics. One of the most widely studied models of protein folding is the hydrophobic-polar (HP) model introduced by Dill. The PFP in the three-dimensional (3D) lattice HP model has been shown to be NP-complete; the proposed algorithms for solving the problem can therefore only find near-optimal energy structures for most long benchmark sequences within acceptable time periods. In this paper, we propose a novel algorithm based on the branch-and-bound approach to solve the PFP in the 3D lattice HP model. For 10 48-monomer benchmark sequences, our proposed algorithm finds the lowest energies so far within comparable computation times than previous methods.

The diversity of Shine-Dalgarno sequences sheds light on the evolution of translation initiation.

Shine-Dalgarno (SD) sequences, the core element of prokaryotic ribosome-binding sites, facilitate mRNA translation by base-pair interaction with the anti-SD (aSD) sequence of 16S rRNA. In contrast to this paradigm, an inspection of thousands of prokaryotic species unravels tremendous SD sequence diversity both within and between genomes, whereas aSD sequences remain largely static. The pattern has led many to suggest unidentified mechanisms for translation initiation. Here we review known translation-initiation pathways in prokaryotes. Moreover, we seek to understand the cause and consequence of SD diversity through surveying recent advances in biochemistry, genomics, and high-throughput genetics. These findings collectively show: (1) SD:aSD base pairing is beneficial but nonessential to translation initiation. (2) The 5' untranslated region of mRNA evolves dynamically and correlates with organismal phylogeny and ecological niches. (3) Ribosomes have evolved distinct usage of translation-initiation pathways in different species. We propose a model portraying the SD diversity shaped by optimization of gene expression, adaptation to environments and growth demands, and the species-specific prerequisite of ribosomes to initiate translation. The model highlights the coevolution of ribosomes and mRNA features, leading to functional customization of the translation apparatus in each organism.

Global fitness landscapes of the Shine-Dalgarno sequence.

Shine-Dalgarno sequences (SD) in prokaryotic mRNA facilitate protein translation by pairing with rRNA in ribosomes. Although conventionally defined as AG-rich motifs, recent genomic surveys reveal great sequence diversity, questioning how SD functions. Here, we determined the molecular fitness (i.e., translation efficiency) of 49 synthetic 9-nt SD genotypes in three distinct mRNA contexts in Escherichia coli We uncovered generic principles governing the SD fitness landscapes: (1) Guanine contents, rather than canonical SD motifs, best predict the fitness of both synthetic and endogenous SD; (2) the genotype-fitness correlation of SD promotes its evolvability by steadily supplying beneficial mutations across fitness landscapes; and (3) the frequency and magnitude of deleterious mutations increase with background fitness, and adjacent nucleotides in SD show stronger epistasis. Epistasis results from disruption of the continuous base pairing between SD and rRNA. This "chain-breaking" epistasis creates sinkholes in SD fitness landscapes and may profoundly impact the evolution and function of prokaryotic translation initiation and other RNA-mediated processes. Collectively, our work yields functional insights into the SD sequence variation in prokaryotic genomes, identifies a simple design principle to guide bioengineering and bioinformatic analysis of SD, and illuminates the fundamentals of fitness landscapes and molecular evolution.

Transhydrogenase promotes the robustness and evolvability of E. coli deficient in NADPH production.

Metabolic networks revolve around few metabolites recognized by diverse enzymes and involved in myriad reactions. Though hub metabolites are considered as stepping stones to facilitate the evolutionary expansion of biochemical pathways, changes in their production or consumption often impair cellular physiology through their system-wide connections. How does metabolism endure perturbations brought immediately by pathway modification and restore hub homeostasis in the long run? To address this question we studied laboratory evolution of pathway-engineered Escherichia coli that underproduces the redox cofactor NADPH on glucose. Literature suggests multiple possibilities to restore NADPH homeostasis. Surprisingly, genetic dissection of isolates from our twelve evolved populations revealed merely two solutions: (1) modulating the expression of membrane-bound transhydrogenase (mTH) in every population; (2) simultaneously consuming glucose with acetate, an unfavored byproduct normally excreted during glucose catabolism, in two subpopulations. Notably, mTH displays broad phylogenetic distribution and has also played a predominant role in laboratory evolution of Methylobacterium extorquens deficient in NADPH production. Convergent evolution of two phylogenetically and metabolically distinct species suggests mTH as a conserved buffering mechanism that promotes the robustness and evolvability of metabolism. Moreover, adaptive diversification via evolving dual substrate consumption highlights the flexibility of physiological systems to exploit ecological opportunities.

Mapping the fitness landscape of gene expression uncovers the cause of antagonism and sign epistasis between adaptive mutations.

How do adapting populations navigate the tensions between the costs of gene expression and the benefits of gene products to optimize the levels of many genes at once? Here we combined independently-arising beneficial mutations that altered enzyme levels in the central metabolism of Methylobacterium extorquens to uncover the fitness landscape defined by gene expression levels. We found strong antagonism and sign epistasis between these beneficial mutations. Mutations with the largest individual benefit interacted the most antagonistically with other mutations, a trend we also uncovered through analyses of datasets from other model systems. However, these beneficial mutations interacted multiplicatively (i.e., no epistasis) at the level of enzyme expression. By generating a model that predicts fitness from enzyme levels we could explain the observed sign epistasis as a result of overshooting the optimum defined by a balance between enzyme catalysis benefits and fitness costs. Knowledge of the phenotypic landscape also illuminated that, although the fitness peak was phenotypically far from the ancestral state, it was not genetically distant. Single beneficial mutations jumped straight toward the global optimum rather than being constrained to change the expression phenotypes in the correlated fashion expected by the genetic architecture. Given that adaptation in nature often results from optimizing gene expression, these conclusions can be widely applicable to other organisms and selective conditions. Poor interactions between individually beneficial alleles affecting gene expression may thus compromise the benefit of sex during adaptation and promote genetic differentiation.

Optimization of gene expression through divergent mutational paths.

Adaptation under similar selective pressure often leads to comparable phenotypes. A longstanding question is whether such phenotypic repeatability entails similar (parallelism) or different genotypic changes (convergence). To better understand this, we characterized mutations that optimized expression of a plasmid-borne metabolic pathway during laboratory evolution of a bacterium. Expressing these pathway genes was essential for growth but came with substantial costs. Starting from overexpression, replicate populations founded by this bacterium all evolved to reduce expression. Despite this phenotypic repetitiveness, the underlying mutational spectrum was highly diverse. Analysis of these plasmid mutations identified three distinct means to modulate gene expression: (1) reducing the gene copy number, (2) lowering transcript stability, and (3) integration of the pathway-bearing plasmid into the host genome. Our study revealed diverse molecular changes beneath convergence to a simple phenotype. This complex genotype-phenotype mapping presents a challenge to inferring genetic evolution based solely on phenotypic changes.

Diminishing returns epistasis among beneficial mutations decelerates adaptation.

Epistasis has substantial impacts on evolution, in particular, the rate of adaptation. We generated combinations of beneficial mutations that arose in a lineage during rapid adaptation of a bacterium whose growth depended on a newly introduced metabolic pathway. The proportional selective benefit for three of the four loci consistently decreased when they were introduced onto more fit backgrounds. These three alleles all reduced morphological defects caused by expression of the foreign pathway. A simple theoretical model segregating the apparent contribution of individual alleles to benefits and costs effectively predicted the interactions between them. These results provide the first evidence that patterns of epistasis may differ for within- and between-gene interactions during adaptation and that diminishing returns epistasis contributes to the consistent observation of decelerating fitness gains during adaptation.

Fast growth increases the selective advantage of a mutation arising recurrently during evolution under metal limitation.

Understanding the evolution of biological systems requires untangling the molecular mechanisms that connect genetic and environmental variations to their physiological consequences. Metal limitation across many environments, ranging from pathogens in the human body to phytoplankton in the oceans, imposes strong selection for improved metal acquisition systems. In this study, we uncovered the genetic and physiological basis of adaptation to metal limitation using experimental populations of Methylobacterium extorquens AM1 evolved in metal-deficient growth media. We identified a transposition mutation arising recurrently in 30 of 32 independent populations that utilized methanol as a carbon source, but not in any of the 8 that utilized only succinate. These parallel insertion events increased expression of a novel transporter system that enhanced cobalt uptake. Such ability ensured the production of vitamin B(12), a cobalt-containing cofactor, to sustain two vitamin B(12)-dependent enzymatic reactions essential to methanol, but not succinate, metabolism. Interestingly, this mutation provided higher selective advantages under genetic backgrounds or incubation temperatures that permit faster growth, indicating growth-rate-dependent epistatic and genotype-by-environment interactions. Our results link beneficial mutations emerging in a metal-limiting environment to their physiological basis in carbon metabolism, suggest that certain molecular features may promote the emergence of parallel mutations, and indicate that the selective advantages of some mutations depend generically upon changes in growth rate that can stem from either genetic or environmental influences.

Asymmetric, bimodal trade-offs during adaptation of Methylobacterium to distinct growth substrates.

Trade-offs between selected and nonselected environments are often assumed to exist during adaptation. This phenomenon is prevalent in microbial metabolism, where many organisms have come to specialize on a narrow breadth of substrates. One well-studied example is methylotrophic bacteria that can use single-carbon (C(1)) compounds as their sole source of carbon and energy, but generally use few, if any, multi-C compounds. Here, we use adaptation of experimental populations of the model methylotroph, Methylobacterium extorquens AM1, to C(1) (methanol) or multi-C (succinate) compounds to investigate specialization and trade-offs between these two metabolic lifestyles. We found a general trend toward trade-offs during adaptation to succinate, but this was neither universal nor showed a quantitative relationship with the extent of adaptation. After 1500 generations, succinate-evolved strains had a remarkably bimodal distribution of fitness values on methanol: either an improvement comparable to the strains adapted on methanol or the complete loss of the ability to grow on C(1) compounds. In contrast, adaptation to methanol resulted in no such trade-offs. Based on the substantial, asymmetric loss of C(1) growth during growth on succinate, we suggest that the long-term maintenance of C(1) metabolism across the genus Methylobacterium requires relatively frequent use of C(1) compounds to prevent rapid loss.