Search

Protein-protein interaction (PPI) networks have been found to be power-law-distributed, \ie, in observed PPI networks, the fraction of nodes with degree koften follows a power law (PL) distribution \(k^{-\alpha}\). It has been hypothesized that, during evolution, this property has emerged due to proteins encoded by paralogs copying the interaction partners of the proteins encoded by the duplicated genes. However, the experimental procedures used to detect PPIs are known to be heavily affected by technical and study bias. For instance, proteins known to be involved in cancer are often heavily overstudied and proteins used as baits in large scale experiments such as yeast two-hybrid (Y2H) or affinity purification-mass spectrometry (AP-MS) tend to have many false-positive interaction partner. This raises the question whether PL distributions in observed PPI networks could be explained by these biases alone. Here, we address this question using statistical analyses of the degree distributions of 1,427 observed PPI networks of controlled provenance as well as simulation studies. Our results indicate that study bias and technical bias can indeed largely explain the fact that observed PPI networks tend to be PL-distributed. This implies that it is problematic to derive hypotheses about the degree distribution and emergence of the true biological interactome from the PL distributions in observed PPI networks.

A key parameter in the experimental design of poly(A) RNA-seq projects is the choice of sequencing depth. Considering a limited budget, one needs to find a tradeoff between the number of samples and the sensitivity of the analysis, in particular with respect to lowly expressed genes. While previous studies have proposed a lower bound for the comprehensive analysis of differential gene expression, for the analysis of alternative splicing has only been proposed for human adipose tissue. Alternative splicing, however, differs across tissues and conditions. We analyzed deeply sequenced paired-end RNA-seq samples (between 150 and >500 million reads, read length 50-150 bp) from human buffy coat cells and diverse tissues, including gluteal subcutaneous fat, heart, and hypothalamus. We provide evidence that for genes with moderate and high expression, 200 million reads is the lower bound for comprehensive analysis of alternative splicing. Hence, given the current state of methodology, integrating splicing into the systems biology landscape requires much deeper sequencing than in most published cohorts.

Additional file 1 Experimental details.Table S1. The SHIP case study. Table S2.. The COPDGene case study. Table S3. The FinnGen case study. Supplementary results.Figure S1. The significant SNPs overlapped between sPLINK and PLINK for the SHIP case study considering Bonferroni significance threshold. Figure S2. The Spearman rank correlation coefficient between the p-values from each tool and the aggregated analysis for the COPDGene and FinnGen case studies. Figure S3. Runtime and network bandwidth usage of sPLINK with varying number of SNPs. Figure S4. Runtime and network bandwidth usage of sPLINK with varying number of samples. Figure S5. Runtime and network bandwidth usage of sPLINK with varying number of clients. Experimental setup.Table S4. The system specification of the physical machines and laptops used to measure the runtime and network bandwidth usage of sPLINK. Table S5. The experimental setup used for measuring the runtime and network bandwidth usage of sPLINK.

Additional file 2 Review history.

Additional file 4 Supplementary Text and Figures S1-S3

Additional file 9 Review history.

Additional file 8. Review history.

Additional file 1: Tables S1-S3. Table S1. Enrichment of the pathway ���Muscle contraction��� from Reactome for the exons upregulated in the muscles generated by the NEASE package. Table S2. Enrichment of the pathway ���Synaptic vesicle cycle��� from KEGG for the exons upregulated in the neural tissues generated by the NEASE package. Table S3. RNA-Seq samples used in the study. Figs. S1-S6. Fig. S1. Node degree distribution of the classic PPI and structurally annotated PPI, the latter contains only interactions with evidence from DDIs and DMIs or residue-level evidence from the co-resolved structure. Fig S2. Network Enrichment Analysis using EviNet webtool for exons upregulated in muscles and neural tissues. Fig. S3. NEASE visualization highlights the interactions of differentially spliced genes with the DCM pathway. Fig. S4. The PSI values of two exon skipping events in the genes TPM1 and DST from the GTEx dataset confirm that both the exons are upregulated in muscles and heart tissues. Fig. S5. The PSI values of two exon skipping events in the genes CLTA and CLTB from the GTEx dataset confirm that both the exons are upregulated in neural tissues. Fig. S6. Pseudocode of NEASE algorithm.

Aggregating transcriptomics data across hospitals can increase sensitivity and robustness of differential expression analyses, yielding deeper clinical insights. As data exchange is often restricted by privacy legislation, meta-analyses are frequently employed to pool local results. However, if class labels are inhomogeneously distributed between cohorts, their accuracy may drop. Flimma (https://exbio.wzw.tum.de/flimma/) addresses this issue by implementing the state-of-the-art workflow limma voom in a privacy-preserving manner, i.e. patient data never leaves its source site. Flimma results are identical to those generated by limma voom on combined datasets even in imbalanced scenarios where meta-analysis approaches fail., Comment: 27 pages, 7 figures

Federated learning has faced performance and network communication challenges, especially in the environments where the data is not independent and identically distributed (IID) across the clients. To address the former challenge, we introduce the federated-centralized concordance property and show that the federated single-mini-batch training approach can achieve comparable performance as the corresponding centralized training in the Non-IID environments. To deal with the latter, we present the federated multi-mini-batch approach and illustrate that it can establish a trade-off between the performance and communication efficiency and outperforms federated averaging in the Non-IID settings.

Artificial intelligence (AI) has been successfully applied in numerous scientific domains. In biomedicine, AI has already shown tremendous potential, e.g. in the interpretation of next-generation sequencing data and in the design of clinical decision support systems. However, training an AI model on sensitive data raises concerns about the privacy of individual participants. For example, summary statistics of a genome-wide association study can be used to determine the presence or absence of an individual in a given dataset. This considerable privacy risk has led to restrictions in accessing genomic and other biomedical data, which is detrimental for collaborative research and impedes scientific progress. Hence, there has been a substantial effort to develop AI methods that can learn from sensitive data while protecting individuals' privacy. This paper provides a structured overview of recent advances in privacy-preserving AI techniques in biomedicine. It places the most important state-of-the-art approaches within a unified taxonomy and discusses their strengths, limitations, and open problems. As the most promising direction, we suggest combining federated machine learning as a more scalable approach with other additional privacy preserving techniques. This would allow to merge the advantages to provide privacy guarantees in a distributed way for biomedical applications. Nonetheless, more research is necessary as hybrid approaches pose new challenges such as additional network or computation overhead., Comment: 17 pages, 3 figures, 3 tables. Methods of Information in Medicine (2022)