Large Language Models (LLMs) are increasingly integral to recommendation systems, offering sophisticated language understanding and generation capabilities. However, their practical application is often hindered by challenges such as data sparsity, the generation of unreliable or hallucinated recommendations, and a general lack of transparency in their decision-making processes. Existing mitigation strategies frequently introduce significant complexity or computational overhead. To address these limitations, particularly the critical gap in quantifying the confidence of LLM-generated recommendations, we propose GUIDER: Uncertainty Guided Dynamic Re-ranking for Large Language Models based Recommender Systems. This new framework innovatively leverages the logits produced by LLMs as evidence for recommended items. By employing a Dirichlet distribution, GUIDER decomposes the total predictive uncertainty into distinct Data Uncertainty (DU), reflecting inherent data ambiguity, and Model Uncertainty (MU), indicating the model's own conviction. This principled decomposition, achieved with a single inference pass, enhances transparency and trustworthiness. Based on the quantified DU and MU levels, our system dynamically adapts its recommendation strategy---adjusting output diversity---through a four-quadrant analysis that tailors responses to specific uncertainty profiles. Extensive experiments conducted in zero-shot recommendation settings validate the effectiveness of our approach. GUIDER consistently outperforms existing methods in reliability-aware scenarios, demonstrably improving recommendation quality. This framework not only advances the practical deployment of LLM-based recommenders by making them more dependable but also provides a robust foundation for future research into uncertainty-aware generative systems.

BACKGROUND: Many centrality measures have been proposed to mine and characterize the correlations between network topological properties and protein essentiality. However, most of them show limited prediction accuracy, and the number of common predicted essential proteins by different methods is very small. RESULTS: In this paper, an ensemble framework is proposed which integrates gene expression data and protein-protein interaction networks (PINs). It aims to improve the prediction accuracy of basic centrality measures. The idea behind this ensemble framework is that different protein-protein interactions (PPIs) may show different contributions to protein essentiality. Five standard centrality measures (degree centrality, betweenness centrality, closeness centrality, eigenvector centrality, and subgraph centrality) are integrated into the ensemble framework respectively. We evaluated the performance of the proposed ensemble framework using yeast PINs and gene expression data. The results show that it can considerably improve the prediction accuracy of the five centrality measures individually. It can also remarkably increase the number of common predicted essential proteins among those predicted by each centrality measure individually and enable each centrality measure to find more low-degree essential proteins. CONCLUSIONS: This paper demonstrates that it is valuable to differentiate the contributions of different PPIs for identifying essential proteins based on network topological characteristics. The proposed ensemble framework is a successful paradigm to this end.

BACKGROUND: There has been a growing interest in identifying context-specific active protein-protein interaction (PPI) subnetworks through integration of PPI and time course gene expression data. However the interaction dynamics during the biological process under study has not been sufficiently considered previously. METHODS: Here we propose a topology-phase locking (TopoPL) based scoring metric for identifying active PPI subnetworks from time series expression data. First the temporal coordination in gene expression changes is evaluated through phase locking analysis; The results are subsequently integrated with PPI to define an activity score for each PPI subnetwork, based on individual member expression, as well topological characteristics of the PPI network and of the expression temporal coordination network; Lastly, the subnetworks with the top scores in the whole PPI network are identified through simulated annealing search. RESULTS: Application of TopoPL to simulated data and to the yeast cell cycle data showed that it can more sensitively identify biologically meaningful subnetworks than the method that only utilizes the static PPI topology, or the additive scoring method. Using TopoPL we identified a core subnetwork with 49 genes important to yeast cell cycle. Interestingly, this core contains a protein complex known to be related to arrangement of ribosome subunits that exhibit extremely high gene expression synchronization. CONCLUSIONS: Inclusion of interaction dynamics is important to the identification of relevant gene networks.

BACKGROUND: Bayesian Network (BN) is a powerful approach to reconstructing genetic regulatory networks from gene expression data. However, expression data by itself suffers from high noise and lack of power. Incorporating prior biological knowledge can improve the performance. As each type of prior knowledge on its own may be incomplete or limited by quality issues, integrating multiple sources of prior knowledge to utilize their consensus is desirable. RESULTS: We introduce a new method to incorporate the quantitative information from multiple sources of prior knowledge. It first uses the Naïve Bayesian classifier to assess the likelihood of functional linkage between gene pairs based on prior knowledge. In this study we included cocitation in PubMed and schematic similarity in Gene Ontology annotation. A candidate network edge reservoir is then created in which the copy number of each edge is proportional to the estimated likelihood of linkage between the two corresponding genes. In network simulation the Markov Chain Monte Carlo sampling algorithm is adopted, and samples from this reservoir at each iteration to generate new candidate networks. We evaluated the new algorithm using both simulated and real gene expression data including that from a yeast cell cycle and a mouse pancreas development/growth study. Incorporating prior knowledge led to a ~2 fold increase in the number of known transcription regulations recovered, without significant change in false positive rate. In contrast, without the prior knowledge BN modeling is not always better than a random selection, demonstrating the necessity in network modeling to supplement the gene expression data with additional information. CONCLUSION: our new development provides a statistical means to utilize the quantitative information in prior biological knowledge in the BN modeling of gene expression data, which significantly improves the performance.

Abstract Summary: Extracting biological insight from microarray data is important but challenging. Here we describe TAPPA, a java-based tool, for identification of phenotype-associated genetic pathways utilizing the pathway topological measures. This is achieved by first calculating a Pathway Connectivity Index (PCI) for each pathway, followed by evaluating its correlation to the phenotypic variation. Our PCI definition not only efficiently captures the contributions from genes that show subtle but consistent changes in expression, but also naturally overweighs the hub genes that interact with a large number of other genes in the pathway. TAPPA also allows evaluation of sub-modules within a pathway and their association to phenotypes. Availability: TAPPA and data for Figure 1 are freely available from http://watson.mcgee.mcw.edu:8080/~sgao Contact: [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.

BACKGROUND: Gene expression profiling using microarrays has become an important genetic tool. Spotted arrays prepared in academic labs have the advantage of low cost and high design and content flexibility, but are often limited by their susceptibility to quality control (QC) issues. Previously, we have reported a novel 3-color microarray technology that enabled array fabrication QC. In this report we further investigated its advantage in spot-level data QC. RESULTS: We found that inadequate amount of bound probes available for hybridization led to significant, gene-specific compression in ratio measurements, increased data variability, and printing pin dependent heterogeneities. The impact of such problems can be captured through the definition of quality scores, and efficiently controlled through quality-dependent filtering and normalization. We compared gene expression measurements derived using our data processing pipeline with the known input ratios of spiked in control clones, and with the measurements by quantitative real time RT-PCR. In each case, highly linear relationships (R2 > 0.94) were observed, with modest compression in the microarray measurements (correction factor < 1.17). CONCLUSION: Our microarray analytical and technical advancements enabled a better dissection of the sources of data variability and hence a more efficient QC. With that highly accurate gene expression measurements can be achieved using the cDNA microarray technology.

Data preprocessing including proper normalization and adequate quality control before complex data mining is crucial for studies using the cDNA microarray technology. We have developed a simple procedure that integrates data filtering and normalization with quantitative quality control of microarray experiments. Previously we have shown that data variability in a microarray experiment can be very well captured by a quality score q(com) that is defined for every spot, and the ratio distribution depends on q(com). Utilizing this knowledge, our data-filtering scheme allows the investigator to decide on the filtering stringency according to desired data variability, and our normalization procedure corrects the q(com)-dependent dye biases in terms of both the location and the spread of the ratio distribution. In addition, we propose a statistical model for false positive rate determination based on the design and the quality of a microarray experiment. The model predicts that a lower limit of 0.5 for the replicate concordance rate is needed in order to be certain of true positives. Our work demonstrates the importance and advantages of having a quantitative quality control scheme for microarrays.