Model Pruning Enables Efficient Federated Learning on Edge Devices.

Federated learning (FL) allows model training from local data collected by edge/mobile devices while preserving data privacy, which has wide applicability to image and vision applications. A challenge is that client devices in FL usually have much more limited computation and communication resources compared to servers in a data center. To overcome this challenge, we propose PruneFL -a novel FL approach with adaptive and distributed parameter pruning, which adapts the model size during FL to reduce both communication and computation overhead and minimize the overall training time, while maintaining a similar accuracy as the original model. PruneFL includes initial pruning at a selected client and further pruning as part of the FL process. The model size is adapted during this process, which includes maximizing the approximate empirical risk reduction divided by the time of one FL round. Our experiments with various datasets on edge devices (e.g., Raspberry Pi) show that: 1) we significantly reduce the training time compared to conventional FL and various other pruning-based methods and 2) the pruned model with automatically determined size converges to an accuracy that is very similar to the original model, and it is also a lottery ticket of the original model.

Enhanced prediction of Src homology 2 (SH2) domain binding potentials using a fluorescence polarization-derived c-Met, c-Kit, ErbB, and androgen receptor interactome.

Many human diseases are associated with aberrant regulation of phosphoprotein signaling networks. Src homology 2 (SH2) domains represent the major class of protein domains in metazoans that interact with proteins phosphorylated on the amino acid residue tyrosine. Although current SH2 domain prediction algorithms perform well at predicting the sequences of phosphorylated peptides that are likely to result in the highest possible interaction affinity in the context of random peptide library screens, these algorithms do poorly at predicting the interaction potential of SH2 domains with physiologically derived protein sequences. We employed a high throughput interaction assay system to empirically determine the affinity between 93 human SH2 domains and phosphopeptides abstracted from several receptor tyrosine kinases and signaling proteins. The resulting interaction experiments revealed over 1000 novel peptide-protein interactions and provided a glimpse into the common and specific interaction potentials of c-Met, c-Kit, GAB1, and the human androgen receptor. We used these data to build a permutation-based logistic regression classifier that performed considerably better than existing algorithms for predicting the interaction potential of several SH2 domains.

Comprehensive binary interaction mapping of SH2 domains via fluorescence polarization reveals novel functional diversification of ErbB receptors.

First-generation interaction maps of Src homology 2 (SH2) domains with receptor tyrosine kinase (RTK) phosphosites have previously been generated using protein microarray (PM) technologies. Here, we developed a large-scale fluorescence polarization (FP) methodology that was able to characterize interactions between SH2 domains and ErbB receptor phosphosites with higher fidelity and sensitivity than was previously achieved with PMs. We used the FP assay to query the interaction of synthetic phosphopeptides corresponding to 89 ErbB receptor intracellular tyrosine sites against 93 human SH2 domains and 2 phosphotyrosine binding (PTB) domains. From 358,944 polarization measurements, the affinities for 1,405 unique biological interactions were determined, 83% of which are novel. In contrast to data from previous reports, our analyses suggested that ErbB2 was not more promiscuous than the other ErbB receptors. Our results showed that each receptor displays unique preferences in the affinity and location of recruited SH2 domains that may contribute to differences in downstream signaling potential. ErbB1 was enriched versus the other receptors for recruitment of domains from RAS GEFs whereas ErbB2 was enriched for recruitment of domains from tyrosine and phosphatidyl inositol phosphatases. ErbB3, the kinase inactive ErbB receptor family member, was predictably enriched for recruitment of domains from phosphatidyl inositol kinases and surprisingly, was enriched for recruitment of domains from tyrosine kinases, cytoskeletal regulatory proteins, and RHO GEFs but depleted for recruitment of domains from phosphatidyl inositol phosphatases. Many novel interactions were also observed with phosphopeptides corresponding to ErbB receptor tyrosines not previously reported to be phosphorylated by mass spectrometry, suggesting the existence of many biologically relevant RTK sites that may be phosphorylated but below the detection threshold of standard mass spectrometry procedures. This dataset represents a rich source of testable hypotheses regarding the biological mechanisms of ErbB receptors.

Targeted protein-omic methods are bridging the gap between proteomic and hypothesis-driven protein analysis approaches.

While proteomic methods have illuminated many areas of biological protein space, many fundamental questions remain with regard to systems-level relationships between mRNAs, proteins and cell behaviors. While mass spectrometric methods offer a panoramic picture of the relative expression and modification of large numbers of proteins, they are neither optimal for the analysis of predefined targets across large numbers of samples nor for assessing differences in proteins between individual cells or cell compartments. Conversely, traditional antibody-based methods are effective at sensitively analyzing small numbers of proteins across small numbers of conditions, and can be used to analyze relative differences in protein abundance and modification between cells and cell compartments. However, traditional antibody-based approaches are not optimal for analyzing large numbers of protein abundances and modifications across many samples. In this article, we will review recent advances in methodologies and philosophies behind several microarray-based, intermediate-level, 'protein-omic' methods, including a focus on reverse-phase lysate arrays and micro-western arrays, which have been helpful for bridging gaps between large- and small-scale protein analysis approaches and have provided insight into the roles that protein systems play in several biological processes.