MOTIVATION: Antibiotic resistance in Mycobacterium tuberculosis (MTB) poses a significant challenge to global public health. Rapid and accurate prediction of antibiotic resistance can inform treatment strategies and mitigate the spread of resistant strains. In this study, we present a novel approach leveraging large language models (LLMs) to predict antibiotic resistance in MTB (LLMTB). Our model is trained and evaluated on genomic data from 12 185 CRyPTIC isolates and their associated resistance profiles, utilizing natural language processing techniques to capture patterns and mutations linked to resistance. The model's architecture integrates state-of-the-art transformer-based LLMs, enabling the analysis of complex genomic sequences and the extraction of critical features relevant to antibiotic resistance. RESULTS: We evaluate our model's performance using a comprehensive dataset of MTB strains, demonstrating its ability to achieve high performance in predicting resistance to various antibiotics. Unlike traditional machine learning methods, fine-tuning or few-shot learning opens avenues for LLMs to adapt to new or emerging drugs, thereby reducing reliance on extensive data curation. Beyond predictive accuracy, LLMTB uncovers deeper biological insights, identifying critical genes, intergenic regions, and novel resistance mechanisms. This method marks a transformative shift in resistance prediction and offers significant potential for enhancing diagnostic capabilities and guiding personalized treatment plans, ultimately contributing to the global effort to combat tuberculosis and antibiotic resistance. AVAILABILITY AND IMPLEMENTATION: All source code is publicly available at https://github.com/ctestagrose/LLMTB.

Non-uniform sampling selects samples at irregular intervals for concise signal representation. Prior works on non-uniform sampling predominantly focused on maximizing reconstruction accuracy or optimizing sample size. However, a trade-off exists between the two factors, as increasing the sample size can improve the reconstruction accuracy, but it decreases the bandwidth efficiency. The reverse is true for under-sampling. Thus, it is important to balance the two factors. This motivates us to consider CAISOS, a CArdinalIty conStraint nOn-uniform Sampling problem that aims to select at most k sample points of a given time-varying signal such that the regenerated signal has the maximum reconstruction accuracy. Applications of CAISOS include fixed-rate sampling and signal compression in the time domain, such as in robotics and fixed-rate speech encoders. It proposes an integer linear programming model to address CAISOS and shows that the time complexity of the same increases exponentially with increasing signal size. It further proves that CAISOS is an NP-complete problem by showing that it is both NP and NP-hard by using a polynomial-time reduction from the 0/1 knapsack problem. Thus, the paper proposes a polynomial time heuristic based on the least-cost branch-and-bound approximation to solve CAISOS. Finally, it demonstrates the effectiveness of the proposed approaches through simulation by comparing them with the existing counterparts using various time-varying signals available in multiple databases.

The non-uniform sampling (NUS) techniques utilize the geometrical or structural characteristics of the signal to gather samples at irregular intervals. Previous works on NUS primarily concentrate on maximizing the reconstruction accuracy or optimizing the sample size. Nonetheless, a trade-off exists between these two factors, as increasing the sample size can improve the reconstruction accuracy but simultaneously decreases the bandwidth efficiency of the network. The reverse is true if the signal is under-sampled. Thus, it is essential to establish a balance between the two factors. This motivates us to consider the problem of determining the set of sample points with cardinality less than equal to a given$k$, such that the regenerated signal has the maximum reconstruction accuracy in terms of the percentage root mean square difference between the input and output signals. It is referred to as “CArdinalIty conStraint non-uniform Sampling (CAISOS)”. The paper presents an integer linear programming model for the CAISOS problem, but its time complexity increases exponentially with increasing signal size. To overcome this, the paper also proposes a polynomial-time heuristic approach based on the least cost (LC) branch-and-bound approach, and provides an extensive performance evaluation of both methods.