There is growing interest in applying artificial intelligence (AI) to automate and support complex decision-making tasks. However, it remains unclear how algorithms compare to human judgment in contexts requiring semantic understanding and domain expertise. We examine this in the context of the judge assignment problem, matching submissions to suitably qualified judges. Specifically, we tackled this problem at the Harvard President’s Innovation Challenge, the university’s premier venture competition awarding over $500,000 to student and alumni startups. This setting represents a real-world environment where high-quality judge assignment is essential. We developed an AI-based judge assignment algorithm, the Hybrid Lexical-Semantic Similarity Ensemble (HLSE), and deployed it at the competition. We then evaluated its performance against human expert assignments using blinded match-quality scores from judges on 309 judge-venture pairs. Using a Mann-Whitney U statistic-based test, we found no statistically significant difference in assignment quality between the two approaches (AUC=0.48, p=0.40); on average, algorithmic matches were rated 3.90 and manual matches 3.94 on a 5-point scale, where 5 indicates an excellent match. Furthermore, manual assignments that previously required a full week could be automated in several hours by the algorithm during deployment. These results demonstrate that HLSE achieves human-expert-level matching quality while offering greater scalability and efficiency, underscoring the potential of AI-driven solutions to support and enhance human decision-making for judge assignment in high-stakes settings.

Due to modality heterogeneity, EEG and facial images have distinct data structures, temporal dynamics, and semantic levels, which restricts conventional multimodal techniques and complicates feature alignment and fusion. As a remedy, we propose a dynamic adaptive fusion-based cross-modal emotion recognition model (DMAF-DCCA). An emotion-salience-guided adaptive attention approach initially amplifies significant EEG frequency bands and face micro-expression regions to improve discriminative feature extraction. Second, multimodal features are mapped into a single space by a task-aware heterogeneous feature alignment module with emotion-semantic contrastive learning, which fosters robustness and semantic consistency. Lastly, fusion weights are adaptively assigned depending on modality reliability using a confidence-based dynamic decision-level fusion technique. Arousal and valence accuracy results for the DEAP and MAHNOB-HCI datasets are 99.60% and 99.58%, respectively, with an overall accuracy of 99.60%.

Major accidents at substations are generally caused by a variety of safety hazards. However, the interactions between the various hazards within the substation, which result in a heterogeneous network of relationships, make it difficult for traditional generalised analytical models to accurately predict hidden dangers. To solve this problem, we propose a knowledge graph-based method for predicting substation accident hazards, designed to block potential accident causal pathways. Based on the knowledge graph of substation accident hazards, we enhance the network's understanding of anomalies in the complex substation environment through rule-based reasoning, while also leveraging the strong nonlinear fitting capabilities of neural networks to deeply explore the hidden relationships in the data, enabling the risk prediction of substation hazards and preventing major accidents. We validated our model using data from a provincial grid company over the past five years and achieved a prediction accuracy of 80%. This fully demonstrates the effectiveness of our approach in improving the prediction accuracy of substation hazards, and highlights its significant theoretical and practical value in enhancing the safety and stability of the entire substation system.

Electric power operations involve complex and variable conditions, where physiological signals such as ECG, PPG, EEG, and EDA are essential for health monitoring and abnormality detection. However, these signals are often contaminated by noise—such as industrial frequency interference, EMG artifacts, and baseline drift—hindering accurate analysis. To address this, we propose CBAM-Unet, a multimodal denoising model that integrates a Contextual Contrast Block (CCB) for multi-scale feature extraction and a Channel-Spatial Attention Module (CBAM) for dynamic feature focusing. The CCB captures both local and global noise patterns through coarse and fine contextual branches, while CBAM enhances key signal regions via channel and spatial attention. Experimental results show that CBAM-Unet outperforms Unet and FCN in SNR and RMSE. In the ECG denoising task under power maintenance scenarios, SNR improves from 7.58 dB to 17.91 dB and RMSE drops from 0.2066 to 0.0687. For PPG signals, SNR increases from 22.16 dB to 38.43 dB with RMSE reduced to 0.0532. These results demonstrate the model’s effectiveness in complex noise environments, offering strong support for real-time health monitoring and safety assurance in power operations.

Although genome-wide association studies (GWAS) have identified multiple Alzheimer’s disease (AD) risk genes by many novel machine learning methods, their ability to explain the "missing heritability" remains insufficient. One possible reason is that these methods typically discard single-nucleotide polymorphisms (SNPs) with marginal main effects during the dimensionality reduction process. Another reason could be the reliance on case-control study designs, which are less statistically informative than the quantitative trait (QT). In addition, synapse plays an important role in the pathogenesis of AD, but only a few genes related to synapse have been reported based on these methods. Thus, different approaches are necessary to better understand and explain the "missing heritability" of AD. By using CSF T-tau/Aβ42ratio as QT, this paper performed genome-wide SNP-SNP interactions based on a supervised machine learning method. A total of 467 statistically significant SNP-SNP interaction pairs were found to have marginal main effects, and explained a high-level variance of T-tau/Aβ42. To further mining AD risk genes that related to synapses, the Protein–Protein Interaction (PPI) network construction method was used. As a result, a subnetwork with 5 genes were found, which might have effects on both synaptic and cognitive functions within AD patients.

Flow visualization is essentially a tool to answer domain experts' questions about flow fields using rendered images. Static flow visualization approaches require domain experts to raise their questions to visualization experts, who develop specific techniques to extract and visualize the flow structures of interest. Interactive visualization approaches allow domain experts to ask the system directly through the visual analytic interface, which provides flexibility to support various tasks. However, in practice, the visual analytic interface may require extra learning effort, which often discourages domain experts and limits its usage in real-world scenarios. In this paper, we propose FlowNL, a novel interactive system with a natural language interface. FlowNL allows users to manipulate the flow visualization system using plain English, which greatly reduces the learning effort. We develop a natural language parser to interpret user intention and translate textual input into a declarative language. We design the declarative language as an intermediate layer between the natural language and the programming language specifically for flow visualization. The declarative language provides selection and composition rules to derive relatively complicated flow structures from primitive objects that encode various kinds of information about scalar fields, flow patterns, regions of interest, connectivities, etc. We demonstrate the effectiveness of FlowNL using multiple usage scenarios and an empirical evaluation.

Recently, Low Density Parity-Check (LDPC) codes based on Affine Permutation Matrices (APM) drew lots of attention. Compared with the Quasi-Cyclic LDPC (QC-LDPC) codes, these kinds of codes have some advantages. APM-LDPC codes obtain lower cycle-distributions, minimum hamming distance and greater girth. This paper explains the importance of cyclic distribution by comparing APM-LDPC codes with QC-LDPC codes. Then a particular form of APM-LDPC codes is proposed and researched. The new codes can low down the cycle-distribution to larger extent. In the following research, an effective method, which constructs the new codes with fixed girth, is proposed. Simulations show that the construction method is reasonable and effective. The transmission performances are better than the traditional methods, as well. Finally, the implementation and verification are carried out on FPGA.

MOTIVATION: Long noncoding RNA (lncRNA) has been verified to interact with other biomolecules especially protein-coding genes (PCGs), thus playing essential regulatory roles in life activities and disease development. However, the inner mechanisms of most lncRNA-PCG relationships are still unclear. Our study investigated the characteristics of true lncRNA-PCG relationships and constructed a novel predictor with machine learning algorithms. RESULTS: We obtained the 307 true lncRNA-PCG pairs from database and found that there are significant differences in multiple characteristics between true and random lncRNA-PCG sets. Besides, 3-fold cross-validation and prediction results on independent test sets show the great AUC values of LR, SVM and RF, among which RF has the best performance with average AUC 0.818 for cross-validation, 0.823 and 0.853 for two independent test sets, respectively. In case study, some candidate lncRNA-PCG relationships in colorectal cancer were found and HOTAIR-COMP interaction was specially exemplified. The proportion of the reported pairs in the predicted positive results was significantly higher than that in negative results (P < 0.05). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In the nonbinary low-density parity check (NB-LDPC) codes decoding algorithms, the iterative hard reliability based on majority logic decoding (IHRB-MLGD) algorithm has poor error correction performance. The essential reason is that the hard information is used in the initialization and iterative processes. For the problem of partial loss of information, when the reliability is assigned during initialization, the error correction performance is improved by modifying the assignment of reliability at initialization. The initialization process is determined by the probability of occurrence of the number of erroneous bits in the symbol and the Hamming distance. In addition, the IHRB-MLGD decoding algorithm uses the hard decision in the iterative decoding process. The improved algorithm adds soft decision information in the iterative process, which improves the error correction performance while only slightly increasing the decoding complexity, and improves the reliability accumulation process which makes the algorithm more stable. The simulation results indicate that the proposed algorithm has a better decoding performance than IHRB algorithm.

MOTIVATION: Genome-scale CRISPR/Cas9 system has been a democratized gene editing technique and widely used to investigate gene functions in some biological processes and diseases especially cancers. Aiming to characterize gene aberrations and assess their effects on cancer, we designed a pipeline to identify the essential genes for pan-cancer. METHODS: CRISPR screening data were used to identify the essential genes that were collected from published data and integrated by Robust Rank Aggregation algorithm. Then, hypergeometrics test and random walks with restart (RWR) were used to predict additional essential genes on broader scale. Finally, the expression status and potential roles of these genes were explored based on TCGA portal and regulatory network analysis. RESULTS: We collected 926 samples from 10 CRISPR-based screening studies involving 33 different types of cancer to identify cancer-essential genes, which consists of 799 protein-coding genes (PCGs) and 97 long non-coding RNAs (lncRNAs). Then, we constructed a 'bi-colored' network with both PCGs and lncRNAs and applied it to predict additional essential genes including 495 PCGs and 280 lncRNAs on a broader scale using hypergeometrics test and RWR. After obtaining all essential genes, we further investigated their potential roles in cancer and found that essential genes have higher and more stable expression levels, and are associated with multiple cancer-associated biological processes and survival time. The regulatory network analysis detected two intriguing modules of essential genes participating in the regulation of cell cycle and ribosome biogenesis in cancer. AVAILABILITY AND IMPLEMENTATION: . SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.