Search

Proteins are the building blocks of life, carrying out fundamental functions in biology. In computational biology, an effective protein representation facilitates many important biological quantifications. Most existing protein representation methods are derived from self‐supervised language models designed for text analysis. Proteins, however, are more than linear sequences of amino acids. Here, a multimodal deep learning framework for incorporating ≈1 million protein sequence, structure, and functional annotation (MASSA) is proposed. A multitask learning process with five specific pretraining objectives is presented to extract a fine‐grained protein‐domain feature. Through pretraining, multimodal protein representation achieves state‐of‐the‐art performance in specific downstream tasks such as protein properties (stability and fluorescence), protein‒protein interactions (shs27k/shs148k/string/skempi), and protein‒ligand interactions (kinase, DUD‐E), while achieving competitive results in secondary structure and remote homology tasks. Moreover, a novel optimal‐transport‐based metric with rich geometry awareness is introduced to quantify the dynamic transferability from the pretrained representation to the related downstream tasks, which provides a panoramic view of the step‐by‐step learning process. The pairwise distances between these downstream tasks are also calculated, and a strong correlation between the inter‐task feature space distributions and adaptability is observed. [ABSTRACT FROM AUTHOR]

With the rise of deep learning technology, due to the superior performance of the deep neural network, its application in the digital economy has attracted extensive attention of scholars. Since the beginning of the 21st century, my country's digital economy has developed rapidly, and its universalization and other characteristics have created favorable conditions for the optimal allocation of resources in underdeveloped regions and the exertion of comparative advantages. The digital economy will play a key role in poverty alleviation, promoting coordinated regional development, narrowing regional gaps, and improving the spatial layout of my country's reform and opening up. This paper studies the factors that the digital economy based on deep neural networks has on regional economic growth. Simulation experiment conclusions are as follows: (1) the digital economy of Guizhou, Beijing, Chongqing, Anhui, and Tibet is growing rapidly. The central and western regions are in a period of rapid growth. For the gap between major industrial provinces, the coefficient of variation reached about 1 before 2013, then it declined rapidly, and slowed down, and steadily declined after 2019, indicating that the gap in the digital economy in various regions is narrowing in general. (2) From the national level, the digital economy index coefficient is 1.24, that is, for every 1% increase in digital economy investment, GDP will increase by about 0.24%. The labor force increased by 1% and the GDP increased by about 0.22%. This promotion effect is also very obvious. (3) Judging from the above data, the western region urgently needs to promote the construction of the digital economy and introduce high-tech digital economy talents. The talent effect in the Midwest has a significant effect on GDP. (4) From the perspective of the whole country and other regions, the parameter coefficients and signs have not changed significantly, so the original model is robust, and so the conclusion is desirable. [ABSTRACT FROM AUTHOR]