The prediction of enzyme kinetic parameters is crucial for screening enzymes with high catalytic efficiency and desired characteristics to catalyze natural or non-natural reactions. Data-driven machine learning models have been explored to reduce experimental cost and speed up the enzyme design process. However, the prediction performance is still subject to significant limitations due to the variance in sequence similarity between training and testing datasets. In this work, we introduce DEKP, an integrated deep learning approach enzyme kinetic parameter prediction. It leverages pretrained models of protein sequences and incorporates enhanced graph neural networks that provide comprehensive representation of protein structural features. This novel approach can effectively alleviate the performance degradation caused by sequence similarity variation. Moreover, it provides sensitive detection of changes in catalytic efficiency due to enzyme mutations. Experiments validate that DEKP outperforms existing models in predicting enzyme kinetic parameters. This work is expected to significantly improve the performance of the enzyme screening process and provide a robust tool for enzyme-directed evolution research.

In the digital era, escalating concerns over personal privacy and social security have led to the advocacy for the “right to be forgotten”, a principle that empowers individuals to request the deletion of their personal data from online platforms. Consequently, machine unlearning (MU) has been proposed as a method for targeted data deletion within machine learning models. However, MU encounters difficulties in distributed learning environments, such as federated learning (FL), where direct access to data is restricted. Federated unlearning (FU) has been developed in response, aiming to facilitate the process of data deletion requests from clients within FL frameworks. Despite advancements, FU methods based on approximate unlearning present a risk of potential data breaches, while methods reliant on retraining necessitate either complete or repeated retraining of clients, which is inefficient. Addressing these challenges, we introduce the federated cluster slicing algorithm (FedCSA), a novel FU strategy that achieves precision and efficiency in data unlearning. FedCSA organizes clients into distinct slices based on model deviation values, facilitating targeted retraining of local models upon unlearning requests. This method not only ensures consistency in the independent and identically distributed degree across slices but also improves unlearning efficiency and maintains global model accuracy. Moreover, FedCSA features an adaptive clustering mechanism that autonomously determines the optimal number of slices, optimizing the unlearning process. Our empirical analysis, conducted across the MNIST, Fashion-MNIST, and CIFAR-10 datasets, underscores FedCSA's superior performance. FedCSA exhibits a fourfold increase in unlearning efficiency compared to traditional retraining methods. Furthermore, when juxtaposed with the sharded, isolated, sliced, and aggregated technique, FedCSA demonstrates a 4%–5% enhancement in global model accuracy. These findings corroborate the efficacy of FedCSA.

Synthetic biology is a highly specialized discipline characterized by diverse and heterogeneous knowledge sources. Leveraging large language models (LLMs) to achieve accurate, context-aware domain-specific question answering remains a significant challenge, particularly due to persistent issues such as knowledge gaps and hallucinations. To address this, we propose APMSR (Adaptive Prompt and Multi-Source Retrieval), an optimization strategy that integrates adaptive prompt generation based on question features with dynamic multi-source knowledge retrieval, guided by the LinUCB algorithm. This approach balances exploration and exploitation to enhance the relevance and precision of domain-specific information retrieval. We implemented APMSR in a question-answering system tailored to synthetic biology and evaluated its performance on complex, professional-level queries. Experimental results demonstrate that the APMSR-optimized system achieves up to 93% accuracy on multiple-choice and true/false questions. These improvements in accuracy and robustness highlight the potential of combining LLMs with retrieval-augmented strategies for advanced domain-specific question answering.