Biology subjects

Nowicka, M.

Publications and source records attributed to Nowicka, M..

8 recordsLinked to original sources

bioRxiv2024

Beware of Data Leakage from Protein LLM Pretraining

Pretrained protein language models are becoming increasingly popular as a backbone for protein property inference tasks such as structure prediction or function annotation, accelerating biological research. However, related research oftentimes does not consider the effects of data leakage from pretraining on the actual downstream task, resulting in potentially unrealistic performance estimates. Reported generalization might not necessarily be reproducible for proteins highly dissimilar from the pretraining set. In this work, we measure the effects of data leakage from protein language model pretraining in the domain of protein thermostability prediction. Specifically, we compare two different dataset split strategies: a pretraining-aware split, designed to avoid similarity between pretraining data and the held-out test sets, and a commonly-used naive split, relying on clustering the training data for a downstream task without taking the pretraining data into account. Our experiments suggest that data leakage from language model pretraining shows consistent effects on melting point prediction across all experiments, distorting the measured performance. The source code and our dataset splits are available at https://github.com/tfiedlerdev/pretraining-aware-hotprot.

bioinformatics↗

bioRxiv2023

Cell age, not chronological age, governs the dynamics and longevity of circulating CD4+ memory T cells

Quantifying the kinetics with which memory T cell populations are generated and maintained is essential for identifying the determinants of the duration of immunity. The quality and persistence of circulating CD4+ effector memory (TEM) and central memory (TCM) T cells in mice appear to shift with age, but it is unclear whether these changes are driven by the aging host environment, by cell age effects, or both. Here we address these issues by combining DNA labelling methods, established fate-mapping systems, a novel reporter mouse strain, and mathematical models. Together, these allow us to quantify the dynamics of both young and established circulating memory CD4+ T cell subsets, within both young and old mice. We show that that these cells and their descendents become more persistent the longer they reside within the TCM and TEM pools. This behaviour may limit memory CD4 T cell diversity by skewing TCR repertoires towards clones generated early in life, but may also compensate for functional defects in new memory cells generated in old age. Author summaryOur long-term protection against infections depends in part on the maintenance of diverse populations of memory CD4 T cells, which are made in response to the initial exposure to the pathogen or a vaccine. These cells are not long-lived, but instead are maintained dynamically at a clonal level through loss and division. Understanding how immune memory persists therefore requires measuring these rates of these processes, and how they might change with age. Here we combine experiments in mice with mathematical models to show that memory CD4 T cells exhibit complex dynamics but increase their capacity to survive as they age. This dynamic implies that as individuals age, their memory CD4 T cell populations become enriched for older clones. This established memory may compensate for functional defects in new T cell responses generated later in life.

immunology↗

bioRxiv2023

Evaluating Prompt Tuning for Conditional Protein Sequence Generation

Designing artificial proteins with specialized functions promises new solutions for biological, medical, and environmental use cases. This field benefits from advances in natural language processing, with state-of-the-art text generation models already being successfully applied to protein sequences. Openly available pre-trained protein language models are able to generate artificial protein sequences and can be finetuned on very specific tasks. Considering the high computational cost of finetuning a model exclusively for one downstream task, prompt tuning has been proposed as a more cost-efficient alternative that shares one model across different tasks. However, no openly available implementation of this approach compatible with protein language models has been previously published. Thus, we adapt an open-source codebase designed for NLP models to build a pipeline for prompt tuning on protein sequence data, supporting the protein language models ProtGPT2 and RITA. We benchmark this implementation for generating proteins of a specific family and evaluate the approach using text processing metrics as well as family membership prediction and protein activity prediction of generated sequences. Our results confirm the advantages of prompt tuning in resource usage, especially storage, encouraging further research and expansion of this technique to related use cases. For our evaluated use case, prompt tuning does not reach up to finetuning in terms of the quality of generated protein sequences, indicating the need for more extensive optimization. Lastly, we observe discrepancies between results of similar evaluation tools, highlighting open problems for principled assessment of protein sequence generation quality.

bioinformatics↗

bioRxiv2021

Pathogenic potential prediction for novel fungal DNA based on a curated fungi-hosts data collection

BackgroundEmerging pathogens are a growing threat, but large data collections and approaches for predicting the risk associated with novel agents are limited to bacteria and viruses. Pathogenic fungi, which also pose a constant threat to public health, remain understudied. Relevant data remains comparatively scarce and scattered among many different sources, hindering the development of sequencing-based detection workflows for novel fungal pathogens. No prediction method working for agents across all three groups is available, even though the cause of an infection is often difficult to identify from symptoms alone. ResultsWe present a curated collection of fungal host range data, comprising records on human, animal and plant pathogens, as well as other plant-associated fungi, linked to publicly available genomes. We show that it can be used to predict the pathogenic potential of novel fungal species directly from DNA sequences with either sequence homology or deep learning. We develop learned, numerical representations of the collected genomes and visualize the landscape of fungal pathogenicity. Finally, we train multi-class models predicting if next-generation sequencing reads originate from novel fungal, bacterial or viral threats. ConclusionsThe neural networks trained using our data collection enable accurate detection of novel fungal pathogens. A curated set of over 1,400 genomes with host and pathogenicity metadata supports training of machine learning models and sequence comparison, not limited to the pathogen detection task. AvailabilityThe data, models and code are hosted at https://zenodo.org/record/5846345, https://zenodo.org/record/5711877, and https://gitlab.com/dacs-hpi/deepac.

bioinformatics↗

bioRxiv2020

A framework for designing miRNA-based distributed cell classifier circuits

MotivationCell classifiers are synthetic bio-devices performing type-specific in vivo classification. The circuits identify a cell state based on its molecular fingerprint. In particular, the classifiers may be designed to recognize cancerous cells and trigger their apoptosis, shaping a novel therapy for cancer patients. Recently, we introduced a new theoretical design of such devices employing distributed classifiers. Here, a group of single-circuit classifiers decides collectively according to a pre-defined threshold function whether a cell is cancerous. The multi-circuit architecture has shown the potential to predict the cell condition with high accuracy. However, lack of far-reaching machinery to design and evaluate distributed cell classifiers, in particular, assessing their robustness to noise and novel information, makes their application limited. ResultsIn this study, we present a comprehensive framework for designing and evaluating miRNA-based distributed cell classifiers comprising data simulation, pre-processing, and an extensive testing scheme. We develop optimization criteria that allow increasing the accuracy and robustness of classifiers to noise and novel information as shown in simulated and real-world data studies. The evaluation performed on cancer data demonstrates that distributed classifiers outperform single-circuit designs in terms of prediction accuracy. Our classifiers include relevant miRNAs previously described in the literature, as well as more complex regulation patterns included in the data. AvailabilityThe code and data are available at: https://github.com/MelaniaNowicka/RAccoon. Contactm.nowicka@fu-berlin.de

bioinformatics↗

bioRxiv2019

Designing Distributed Cell Classifier Circuits using a Genetic Algorithm

Cell classifiers are decision-making synthetic circuits that allow in vivo cell-type classification. Their design is based on finding a relationship between differential expression of miRNAs and the cell condition. Such biological devices have shown potential to become a valuable tool in cancer treatment as a new type-specific cell targeting approach. So far, only single-circuit classifiers were designed in this context. However, reliable designs come with high complexity, making them difficult to assemble in the lab. Here, we apply so-called Distributed Classifiers (DC) consisting of simple single circuits, that decide collectively according to a threshold function. Such architecture potentially simplifies the assembly process and provides design flexibility. Here, we present a genetic algorithm that allows the design and optimization of DCs. Breast cancer case studies show that DCs perform with high accuracy on real-world data. Optimized classifiers capture biologically relevant miRNAs that are cancer-type specific. The comparison to a single-circuit classifier design approach shows that DCs perform with significantly higher accuracy than individual circuits. The algorithm is implemented as an open source tool.

bioinformatics↗

bioRxiv2019

Differential impact of self and environmental antigens on the ontogeny and maintenance of CD4+ T cell memory

Laboratory mice develop populations of circulating memory CD4+ T cells in the absence of overt infection. We have previously shown that these populations are replenished from naive precursors at high levels throughout life (Gossel et al., 2017). However, the nature, relative importance and timing of the forces generating these cells remain unclear. Here, we tracked the generation of memory CD4+ T cell subsets in mice housed in facilities differing in their dirtiness. We found evidence for sequential naive to central memory to effector memory development, and confirmed that both memory subsets are heterogeneous in their rates of turnover. We also inferred that early exposure to self and environmental antigens establishes persistent memory populations at levels determined largely, though not exclusively, by the dirtiness of the environment. After the first few weeks of life, however, these populations are continuously supplemented by new memory cells at rates that are independent of environment.

immunology↗

bioRxiv2018

diffcyt: Differential discovery in high-dimensional cytometry via high-resolution clustering

1High-dimensional flow and mass cytometry allow cell types and states to be characterized in great detail by measuring expression levels of more than 40 targeted protein markers per cell. Here, we present diffcyt, a new computational framework for differential discovery analyses in these datasets, based on (i) high-resolution clustering and (ii) empirical Bayes moderated tests adapted from transcriptomics. Our approach provides improved statistical performance, including for rare cell populations, along with flexible experimental designs and fast runtimes in an open-source framework.

bioinformatics↗