UniDL4BioPep: a universal deep learning architecture for binary classification in peptide bioactivity.

Identification of potent peptides through model prediction can reduce benchwork in wet experiments. However, the conventional process of model buildings can be complex and time consuming due to challenges such as peptide representation, feature selection, model selection and hyperparameter tuning. Recently, advanced pretrained deep learning-based language models (LMs) have been released for protein sequence embedding and applied to structure and function prediction. Based on these developments, we have developed UniDL4BioPep, a universal deep-learning model architecture for transfer learning in bioactive peptide binary classification modeling. It can directly assist users in training a high-performance deep-learning model with a fixed architecture and achieve cutting-edge performance to meet the demands in efficiently novel bioactive peptide discovery. To the best of our best knowledge, this is the first time that a pretrained biological language model is utilized for peptide embeddings and successfully predicts peptide bioactivities through large-scale evaluations of those peptide embeddings. The model was also validated through uniform manifold approximation and projection analysis. By combining the LM with a convolutional neural network, UniDL4BioPep achieved greater performances than the respective state-of-the-art models for 15 out of 20 different bioactivity dataset prediction tasks. The accuracy, Mathews correlation coefficient and area under the curve were 0.7-7, 1.23-26.7 and 0.3-25.6% higher, respectively. A user-friendly web server of UniDL4BioPep for the tested bioactivities is established and freely accessible at https://nepc2pvmzy.us-east-1.awsapprunner.com. The source codes, datasets and templates of UniDL4BioPep for other bioactivity fitting and prediction tasks are available at https://github.com/dzjxzyd/UniDL4BioPep.

Coverage-Balancing User Selection in Mobile Crowd Sensing with Budget Constraint.

Mobile crowd sensing (MCS) is a new computing paradigm for the internet of things, and it is widely accepted as a powerful means to achieve urban-scale sensing and data collection. In the MCS campaign, the smart mobilephone users can detect their surrounding environments with their on-phone sensors and return the sensing data to the MCS organizer. In this paper, we focus on the coverage-balancing user selection (CBUS) problem with a budget constraint. Solving the CBUS problem aims to select a proper subset of users such that their sensing coverage is as large and balancing as possible, yet without violating the budget specified by the MCS campaign. We first propose a novel coverage balance-based sensing utility model, which effectively captures the joint requirement of the MCS requester for coverage area and coverage balance. We then formally define the CBUS problem under the proposed sensing utility model. Because of the NP-hardness of the CBUS problem, we design a heuristic-based algorithm, called MIA, which tactfully employs the maximum independent set model to determine a preliminary subset of users from all the available users and then adjusts this user subset to improve the budget implementation. MIA also includes a fast approach to calculating the area of the union coverage with any complicated boundaries, which is also applicable to any MCS scenarios that are set up with the coverage area-based sensing utility. The extensive numeric experiments show the efficacy of our designs both in coverage balance and in the total coverage area.

Link Investigation of IEEE 802.15.4 Wireless Sensor Networks in Forests.

Wireless sensor networks are expected to automatically monitor the ecological evolution and wildlife habits in forests. Low-power links (transceivers) are often adopted in wireless sensor network applications, in order to save the precious sensor energy and then achieve long-term, unattended monitoring. Recent research has presented some performance characteristics of such low-power wireless links under laboratory or outdoor scenarios with less obstacles, and they have found that low-power wireless links are unreliable and prone to be affected by the target environment. However, there is still less understanding about how well the low-power wireless link performs in real-world forests and to what extent the complex in-forest surrounding environments affect the link performances. In this paper, we empirically evaluate the low-power links of wireless sensors in three typical different forest environments. Our experiment investigates the performance of the link layer compatible with the IEEE 802.15.4 standard and analyzes the variation patterns of the packet reception ratio (PRR), the received signal strength indicator (RSSI) and the link quality indicator (LQI) under diverse experimental settings. Some observations of this study are inconsistent with or even contradict prior results that are achieved in open fields or relatively clean environments and thus, provide new insights both into effectively evaluating the low-power wireless links and into efficiently deploying wireless sensor network systems in forest environments.