Accurately predicting the mechanisms and properties of potential drug molecules is essential for advancing drug discovery. However, traditional methods often require the development of specialized models for each specific prediction task, resulting in inefficiencies in both model training and integration into workflows. Moreover, these approaches are typically limited to predicting pharmaceutical attributes represented as discrete categories, and struggle with predicting complex attributes that are best described in free-form texts. To address these challenges, we introduce DrugChat, a multi-modal large language model (LLM) designed to provide comprehensive predictions of molecule mechanisms and properties within a unified framework. DrugChat analyzes the structure of an input molecule along with users’ queries to generate comprehensive, free-form predictions on drug indications, pharmacodynamics, and mechanisms of action. Moreover, DrugChat supports multi-turn dialogues with users, facilitating interactive and in-depth exploration of the same molecule. Our extensive evaluation, including assessments by human experts, demonstrates that DrugChat significantly outperforms GPT-4 and other leading LLMs in generating accurate free-form predictions, and exceeds state-of-the-art specialized prediction models.

To verify the accuracy and rationality of discrete fracture network modeling, a comprehensive method for evaluating the similarity between two-dimensional fracture maps is proposed in this study. Five evaluation indicators for two-dimensional discrete fracture networks are proposed, including spacing, length, angle, position, and density of rock fractures. Based on the quantitative characteristics of each indicator, methods for evaluating data distribution similarity based on the Wasserstein distance function and matrix similarity based on vector cosine values are proposed. Considering the grouping characteristics of trace lines that are neglected in existing similarity systems, the similarity between each indicator is calculated based on the grouping matching of trace line maps. This achieves the comprehensive calculation of grouping similarity and overall similarity indicators. The importance and size effects of indicators are discussed through fracture model examples. The results show that the indicators of grouping similarity are more rigorous and suitable for engineering modeling with strict requirements compared to those for overall similarity. The proposed method effectively alleviates the difficulty of comprehensive distribution similarity evaluation and achieves a more objective similarity evaluation of position and density indicators. Finally, the established similarity evaluation system is applied to real tunnel construction in Hebei Province, China, providing application value for the accuracy of fracture modeling in practical engineering.

The invention discloses a device and method for building detection and repair based on an unmanned aerial vehicle. The method comprises the steps of modelling an engineering structure through an unmanned aerial vehicle, and determining the track range of automatic cruise detection of the unmanned aerial vehicle; acquiring data of the unmanned aerial vehicle flying to a to-be-detected constructional engineering structure part in the track range, wherein the data comprises non-contact imaging data and contact imaging data; determining a to-be-repaired structural part and coordinates of the to-be-repaired structural part according to non-contact imaging data and contact imaging data; operating the unmanned aerial vehicle according to the coordinates of the to-be-repaired structural part to go to the coordinates of the to-be-repaired structural part; starting a repairing system of the unmanned aerial vehicle, and repairing the to-be-repaired structure part; and carrying out secondary repair on the repaired structural part. According to the invention, buildings can be detected and repaired quickly, so that the construction period for repairing and maintaining the building is shortened.

A robust error-handling system is critical to any project. It ensures smooth application operation and streamlines debugging during development. This post shares insights from developing an error-handling system for FoxiApply, a product by Lyde. Key takeaways include defining custom errors, implementing a global error-handler, adopting diverse strategies for different error types, and retaining the original error during error encapsulation.

Basic Knowledge about Authentication in Web Development

Integrate Stripe Payment to your Web Application

Implement JobQueue using Bull and Redis in your Web Application

Large language models can be instruction tuned with a set of questions and answers. However, to further fine-tune a model on your own data, you need a large number of questions and answers about your data. Producing those questions and answers can be a lot of manual work. This repository lets you use a non-fine-tuned language model (ChatGPT) to extract question/answer pairs automatically from existing textual data, eliminating all manual work.