Single-cell foundation models such as scGPT represent a significant advancement in single-cell omics, with an ability to achieve state-of-the-art performance on various downstream biological tasks. However, these models are inherently limited in that a vast amount of information in biology exists as text, which they are unable to leverage. There have therefore been several recent works that propose the use of LLMs as an alternative to single-cell foundation models, achieving competitive results. However, there is little understanding of what factors drive this performance, along with a strong focus on using LLMs as an alternative, rather than complementary approach to single-cell foundation models. In this study, we therefore investigate what biological insights contribute toward the performance of LLMs when applied to single-cell data, and introduce scMPT; a model which leverages synergies between scGPT, and single-cell representations from LLMs that capture these insights. scMPT demonstrates stronger, more consistent performance than either of its component models, which frequently have large performance gaps between each other across datasets. We also experiment with alternate fusion methods, demonstrating the potential of combining specialized reasoning models with scGPT to improve performance. This study ultimately showcases the potential for LLMs to complement single-cell foundation models and drive improvements in single-cell analysis.

Pre-trained large language models have significantly improved code generation. As these models scale up, there is an increasing need for the output to handle more intricate tasks and to be appropriately specialized to particular domains. Here, we target bioinformatics due to the amount of specialized domain knowledge, algorithms, and data operations this discipline requires. We present BioCoder, a benchmark developed to evaluate large language models (LLMs) in generating bioinformatics-specific code. BioCoder spans a broad spectrum of the field and covers cross-file dependencies, class declarations, and global variables. It incorporates 1026 Python functions and 1243 Java methods extracted from GitHub, along with 253 examples from the Rosalind Project, all pertaining to bioinformatics. Using topic modeling we show that overall coverage of the included code is representative of the full spectrum of bioinformatics calculations. BioCoder incorporates a fuzz-testing framework for evaluation. We have applied it to evaluate many models including InCoder, CodeGen, CodeGen2, SantaCoder, StarCoder, StarCoder+, InstructCodeT5+, GPT-3.5, and GPT-4. Furthermore, we finetuned StarCoder, demonstrating how our dataset can effectively enhance the performance of LLMs on our benchmark (by >15% in terms of Pass@K in certain prompt configurations and always >3%). The results highlight two key aspects of successful models: (1) Successful models accommodate a long prompt (> ~2600 tokens) with full context, for functional dependencies. (2) They contain specific domain knowledge of bioinformatics, beyond just general coding knowledge. This is evident from the performance gain of GPT-3.5/4 compared to the smaller models on the benchmark (50% vs up to ~25%). Our dataset, benchmark, Docker images, and scripts required for testing are all available at https://github.com/gersteinlab/biocoder.

To interpret the genetic profile present in a patient sample, it is necessary to know which mutations have important roles in the development of the corresponding cancer type. Named entity recognition is a core step in the text mining pipeline which facilitates mining valuable cancer information from the scientific literature. However, due to the scarcity of related datasets, previous NER attempts in this domain either suffer from low performance when deep learning based models are deployed, or they apply feature based machine learning models or rule based models to tackle this problem, which requires intensive efforts from domain experts, and limit the model generalization capability. In this paper, we propose EPICURE, an ensemble pre trained model equipped with a conditional random field pattern layer and a span prediction pattern layer to extract cancer mutations from text. We also adopt a data augmentation strategy to expand our training set from multiple datasets. Experimental results on three benchmark datasets show competitive results compared to the baseline models.

Many natural language processing tasks, such as named entity recognition (NER), part of speech (POS) tagging, word segmentation, and etc., can be formulated as sequential data labeling problems. Building a sound labeler requires very large number of correctly labeled training examples, which may not always be possible. On the other hand, crowdsourcing provides an inexpensive yet efficient alternative to collect manual sequential labeling from non-experts. However the quality of crowd labeling cannot be guaranteed, and three kinds of errors are typical: (1) incorrect annotations due to lack of expertise (e.g., labeling gene names from plain text requires corresponding domain knowledge); (2) ignored or omitted annotations due to carelessness or low confidence; (3) noisy annotations due to cheating or vandalism. To correct these mistakes, we present Sembler, a statistical model for ensembling crowd sequential labelings. Sembler considers three types of statistical information: (1) the majority agreement that proves the correctness of an annotation; (2) correct annotation that improves the credibility of the corresponding annotator; (3) correct annotation that enhances the correctness of other annotations which share similar linguistic or contextual features. We evaluate the proposed model on a real Twitter and a synthetical biological data set, and find that Sembler is particularly accurate when more than half of annotators make mistakes.