Search

• Automatic diatom identification and visualization in a virtual slide. • Effective at identifying all diatoms under very complex observable backgrounds. • Faster diatom testing in a large tissue smear within minutes than manual identification. • Provides evidence of immersion/submersion for drowning diagnosis and site inference in forensic practice. • A series of standards on model optimization and evaluation are first proposed for automated diatom identification. Diatom testing is supportive for drowning diagnosis in forensic medicine. However, it is very time-consuming and labor-intensive for technicians to identify microscopically a handful of diatoms in sample smears, especially under complex observable backgrounds. Recently, we successfully developed a software, named DiatomNet v1.0 intended to automatically identify diatom frustules in a whole slide under a clear background. Here, we introduced this new software and performed a validation study to elucidate how DiatomNet v1.0 improved its performance with the influence of visible impurities. DiatomNet v1.0 has an intuitive, user-friendly and easy-to-learn graphical user interface (GUI) built in the Drupal and its core architecture for slide analysis including a convolutional neural network (CNN) is written in Python language. The build-in CNN model was evaluated for diatom identification under very complex observable backgrounds with mixtures of common impurities, including carbon pigments and sand sediments. Compared to the original model, the enhanced model following optimization with limited new datasets was evaluated systematically by independent testing and random control trials (RCTs). In independent testing, the original DiatomNet v1.0 was moderately affected, especially when higher densities of impurities existed, and achieved a low recall of 0.817 and F1 score of 0.858 but good precision of 0.905. Following transfer learning with limited new datasets, the enhanced version had better results, with recall and F1 score values of 0.968. A comparative study on real slides showed that the upgraded DiatomNet v1.0 obtained F1 scores of 0.86 and 0.84 for carbon pigment and sand sediment, respectively, slightly worse than manual identification (carbon pigment: 0.91; sand sediment: 0.86), but much less time was needed. The study verified that forensic diatom testing with aid of DiatomNet v1.0 is much more efficient than traditionally manual identification even under complex observable backgrounds. In terms of forensic diatom testing, we proposed a suggested standard on build-in model optimization and evaluation to strengthen the software's generalization in potentially complex conditions. [ABSTRACT FROM AUTHOR]

Background: RNAs play key roles in cells through the interactions with proteins known as the RNA-binding proteins (RBP) and their binding motifs enable crucial understanding of the post-transcriptional regulation of RNAs. How the RBPs correctly recognize the target RNAs and why they bind specific positions is still far from clear. Machine learning-based algorithms are widely acknowledged to be capable of speeding up this process. Although many automatic tools have been developed to predict the RNA-protein binding sites from the rapidly growing multi-resource data, e.g. sequence, structure, their domain specific features and formats have posed significant computational challenges. One of current difficulties is that the cross-source shared common knowledge is at a higher abstraction level beyond the observed data, resulting in a low efficiency of direct integration of observed data across domains. The other difficulty is how to interpret the prediction results. Existing approaches tend to terminate after outputting the potential discrete binding sites on the sequences, but how to assemble them into the meaningful binding motifs is a topic worth of further investigation. Results: In viewing of these challenges, we propose a deep learning-based framework (iDeep) by using a novel hybrid convolutional neural network and deep belief network to predict the RBP interaction sites and motifs on RNAs. This new protocol is featured by transforming the original observed data into a high-level abstraction feature space using multiple layers of learning blocks, where the shared representations across different domains are integrated. To validate our iDeep method, we performed experiments on 31 large-scale CLIP-seq datasets, and our results show that by integrating multiple sources of data, the average AUC can be improved by 8% compared to the best single-source-based predictor; and through cross-domain knowledge integration at an abstraction level, it outperforms the state-of-the-art predictors by 6%. Besides the overall enhanced prediction performance, the convolutional neural network module embedded in iDeep is also able to automatically capture the interpretable binding motifs for RBPs. Large-scale experiments demonstrate that these mined binding motifs agree well with the experimentally verified results, suggesting iDeep is a promising approach in the real-world applications. Conclusion: The iDeep framework not only can achieve promising performance than the state-of-the-art predictors, but also easily capture interpretable binding motifs. iDeep is available at http://www.csbio.sjtu.edu.cn/bioinf/iDeep [ABSTRACT FROM AUTHOR]