Proteins are the foundation of life, and designing functional proteins remains a key challenge in biotechnology. Before the development of AlphaFold2, the focus of design was primarily on structure-centric approaches such as using the well-known open-source software Rosetta3. Following the development of AlphaFold2, deep-learning techniques for protein design gained prominence. This study proposes a new method to generate functional proteins using the diffusion model and ESM2 protein language model. Diffusion models, which are widely used in image and natural language generation, are used here for protein design, facilitating the controlled generation of new sequences. The ESM2 model, trained on the basis of large-scale protein sequence data, provides a deep understanding of the context of the sequence, thus improving the model’s ability to generate biologically relevant proteins. In this study, we used the Protein A-like peptide as a model study object, combined the diffusion model and the ESM2 model to generate new peptide sequences from minimal input data, and verified their biological activities through experiments such as the BLI affinity test. In conclusion, we developed a new method for protein design that provides a novel strategy to meet the challenges of generic protein generation.

In this study, an ultrasonic-assisted procedure for the extraction of mulberry leaf polysaccharides (MLPs) was investigated using response surface methodology with a 29-run Box–Behnken design. Four factors were investigated, and it was found that the factors influencing the process, in order of significance, were the extraction temperature > liquid-to-material ratio > ultrasonic power. Considering practical conditions, the parameters were adjusted to a liquid-to-material ratio of 16:1 mL/g, extraction time of 58 min, extraction temperature of 65 °C, and ultrasonic power of 500 W. Under these conditions, the yield of MLPs was 14.47%, which is close to the predicted value, indicating that the extraction process optimized by response surface methodology (RSM) is feasible. The separation and purification effects of macroporous resin and activated carbon on MLPs were investigated, with the D152 resin being considered the most suitable choice. The optimal separation conditions were found to be a sample concentration of 0.5 g/mL and an optimal flow rate of 1 mL/min. Thin-layer chromatography and infrared spectroscopy revealed that polysaccharides extracted from mulberry leaves are primarily composed of rhamnose, xylose, and arabinose. In conclusion, this study successfully optimized the ultrasonic-assisted extraction process of MLPs through response surface methodology, determined the optimal parameter combination, and verified its efficiency and stability. Under the optimal conditions obtained for ultrasonic-assisted extraction, the yield of MLPs is significantly higher than that reported in the previous literature. The decolorization process of crude mulberry leaf polysaccharide extract was also investigated, and the purified MLPs have clear monosaccharide composition and structural characteristics, providing a theoretical basis and technical support for their application in functional food or drug development.

Due to its high sensitivity and non-destructive nature, Raman spectroscopy has become an essential analytical tool in biopharmaceutical analysis and drug development. Despite of the computational demands, data requirements, or ethical considerations, artificial intelligence (AI) and particularly deep learning algorithms has further advanced Raman spectroscopy by enhancing data processing, feature extraction, and model optimization, which not only improves the accuracy and efficiency of Raman spectroscopy detection, but also greatly expands its range of application. AI-guided Raman spectroscopy has numerous applications in biomedicine, including characterizing drug structures, analyzing drug forms, controlling drug quality, identifying components, and studying drug-biomolecule interactions. AI-guided Raman spectroscopy has also revolutionized biomedical research and clinical diagnostics, particularly in disease early diagnosis and treatment optimization. Therefore, AI methods are crucial to advancing Raman spectroscopy in biopharmaceutical research and clinical diagnostics, offering new perspectives and tools for disease treatment and pharmaceutical process control. In summary, integrating AI and Raman spectroscopy in biomedicine has significantly improved analytical capabilities, offering innovative approaches for research and clinical applications.

Online Raman spectroscopy technology uses frequency shift of scattered light to obtain real-time characteristic peak information of the test substance, combined with machine learning regression models for real-time concentration prediction. However, in the microbial cultivation process, due to the complex and variable composition of the cultivation system substances, this poses a great challenge to the performance of the model. In addition, model training requires a large amount of data, and the process of collecting a large number of samples for offline detection is very difficult. In this study, a convolutional neural network was introduced to replace the partial least squares model in chemometrics to improve the accuracy of model prediction, and a non-offline sampling labeling technique was proposed to obtain a semi-supervised training set with sufficient data. For multiple detection substances, each substance's feature peak position was optimized using a genetic algorithm, and transfer learning was implemented through fine-tuning model structure and hyperparameters to retrain difficult-to-sample substance correlation models. Introduction of transfer learning also adapts to errors in the production scaling-up process without the need for completely retraining the model. With its unique convolutional kernel structure, the convolutional neural network replaces the tedious sliding window filtering and calibration processes in spectral pre-processing, simplifying the spectral processing workflow. After multiple batches of data cross-validation training, the model exhibits excellent robustness and accuracy, and is suitable for predicting substances in microbial cultivation process.