TLCrys: Transfer Learning Based Method for Protein Crystallization Prediction

X-ray diffraction technique is one of the most common methods of ascertaining protein structures, yet only 2–10% of proteins can produce diffraction-quality crystals. Several computational methods have been proposed so far to predict protein crystallization. Nevertheless, the current state-of-the-art computational methods are limited by the scarcity of experimental data. Thus, the prediction accuracy of existing models hasn’t reached the ideal level. To address the problems above, we propose a novel transfer-learning-based framework for protein crystallization prediction, named TLCrys. The framework proceeds in two steps: pre-training and fine-tuning. The pre-training step adopts attention mechanism to extract both global and local information of the protein sequences. The representation learned from the pre-training step is regarded as knowledge to be transferred and fine-tuned to enhance the performance of crystalization prediction. During pre-training, TLCrys adopts a multi-task learning method, which not only improves the learning ability of protein encoding, but also enhances the robustness and generalization of protein representation. The multi-head self-attention layer guarantees that different levels of the protein representation can be extracted by the fine-tuned step. During transfer learning, the fine-tuning strategy used by TLCrys improves the task-specialized learning ability of the network. Our method outperforms all previous predictors significantly in five crystallization stages of prediction. Furthermore, the proposed methodology can be well generalized to other protein sequence classification tasks.

Scellseg: a style-aware cell instance segmentation tool with pre-training and contrastive fine-tuning

Deep learning-based cell segmentation is increasingly utilized in cell biology and molecular pathology, due to massive accumulation of diverse large-scale datasets and excellent performance in cell representation. However, the development of specialized algorithms has long been hampered by a paucity of annotated training data, whereas the performance of generalist algorithm was limited without experiment-specific calibration. Here, we present a deep learning-based tool called Scellseg consisted of novel pre-trained network architecture and contrastive fine-tuning strategy. In comparison to four commonly used algorithms, Scellseg outperformed in average precision on three diverse datasets with no need for dataset-specific configuration. Interestingly, we found that eight images are sufficient for model tuning to achieve satisfied performance based on a shot data scale experiment. We also developed a graphical user interface integrated with functions of annotation, fine-tuning and inference, that allows biologists to easily specialize their own segmentation model and analyze data at the single-cell level.

Adjustable Speed Control and Damping Analysis of Torsional Vibrations in VSD Compressor Systems

This paper proposes a model-based two-degree-of-freedom (2DOF) speed control for a medium voltage (MV) variable speed drive (VSD) connected to a centrifugal compressor (CC) train. Torsional mode excitations in the drive shaft due to converter switching behaviour are considered. An effective description of the harmonics transfer is proposed. The tuning strategy aims to optimize the tracking behaviour of the step and ramp command, taking care of critical speed excitations. The stability of the closed-loop dynamics against time delay and drive parameter variations are studied by means of Nyquist diagrams and time-domain simulations. A descriptive method for the process damping behaviour is proposed. The control strategy is evaluated through simulations as well as an experimental setup, based on a hardware in the loop (HIL) in a master–slave configuration.

Prediction of COVID-19 epidemic situation via fine-tuned IndRNN

The COVID-19 pandemic is the most serious catastrophe since the Second World War. To predict the epidemic more accurately under the influence of policies, a framework based on Independently Recurrent Neural Network (IndRNN) with fine-tuning are proposed for predict the epidemic development trend of confirmed cases and deaths in the United Stated, India, Brazil, France, Russia, China, and the world to late May, 2021. The proposed framework consists of four main steps: data pre-processing, model pre-training and weight saving, the weight fine-tuning, trend predicting and validating. It is concluded that the proposed framework based on IndRNN and fine-tuning with high speed and low complexity, has great fitting and prediction performance. The applied fine-tuning strategy can effectively reduce the error by up to 20.94% and time cost. For most of the countries, the MAPEs of fine-tuned IndRNN model were less than 1.2%, the minimum MAPE and RMSE were 0.05%, and 1.17, respectively, by using Chinese deaths, during the testing phase. According to the prediction and validation results, the MAPEs of the proposed framework were less than 6.2% in most cases, and it generated lowest MAPE and RMSE values of 0.05% and 2.14, respectively, for deaths in China. Moreover, Policies that play an important role in the development of COVID-19 have been summarized. Timely and appropriate measures can greatly reduce the spread of COVID-19; untimely and inappropriate government policies, lax regulations, and insufficient public cooperation are the reasons for the aggravation of the epidemic situations. The code is available at https://github.com/zhhongsh/COVID19-Precdiction. And the prediction by IndRNN model with fine-tuning are now available online (http://47.117.160.245:8088/IndRNNPredict).

On-device Prior Knowledge Incorporated Learning for Personalized Atrial Fibrillation Detection

Atrial Fibrillation (AF), one of the most prevalent arrhythmias, is an irregular heart-rate rhythm causing serious health problems such as stroke and heart failure. Deep learning based methods have been exploited to provide an end-to-end AF detection by automatically extracting features from Electrocardiogram (ECG) signal and achieve state-of-the-art results. However, the pre-trained models cannot adapt to each patient’s rhythm due to the high variability of rhythm characteristics among different patients. Furthermore, the deep models are prone to overfitting when fine-tuned on the limited ECG of the specific patient for personalization. In this work, we propose a prior knowledge incorporated learning method to effectively personalize the model for patient-specific AF detection and alleviate the overfitting problems. To be more specific, a prior-incorporated portion importance mechanism is proposed to enforce the network to learn to focus on the targeted portion of the ECG, following the cardiologists’ domain knowledge in recognizing AF. A prior-incorporated regularization mechanism is further devised to alleviate model overfitting during personalization by regularizing the fine-tuning process with feature priors on typical AF rhythms of the general population. The proposed personalization method embeds the well-defined prior knowledge in diagnosing AF rhythm into the personalization procedure, which improves the personalized deep model and eliminates the workload of manually adjusting parameters in conventional AF detection method. The prior knowledge incorporated personalization is feasibly and semi-automatically conducted on the edge, device of the cardiac monitoring system. We report an average AF detection accuracy of 95.3% of three deep models over patients, surpassing the pre-trained model by a large margin of 11.5% and the fine-tuning strategy by 8.6%.

Time-Varying Transmissibility Analysis of Vehicle–Bridge Interaction Systems with Application to Bridge-Friendly Vehicles

It is well known that the natural frequencies of a coupled vehicle–bridge interaction system are time-varying. While this knowledge is useful for applications in bridge health monitoring, it does not provide an understanding of the relations between the excitation and coupled system responses, nor leads to developments of effective control strategies to mitigate vibration. In this paper, a novel theoretical framework for the time-varying displacement transmissibility is developed using a time-frozen technique. The time–frequency characteristics of the transmissibility functions are investigated to gain fundamental understanding and insights of the coupling dynamics in relation to the matching of bridge and vehicle natural frequencies. An important aspect of the transmissibility formulation is that it leads to the development of physics-based vibration control strategies in the frequency domain. By applying the principle of fixed points from vibration absorber designs to the transmissibility functions, an optimally tuned vehicle suspension to mitigate bridge vibration is obtained. The tuning strategy depends only on a priori known structural parameters. Thus, the tuning strategy provides useful guidelines in practice and is shown to be effective in reducing the vibrations of both the moving vehicle and the bridge. This work paves a foundation for further research in the design of bridge-friendly vehicles via parameter tuning.