Biology subjects

Basham, M.

Publications and source records attributed to Basham, M..

11 recordsLinked to original sources

bioRxiv2026

Adaptive Milling: imaging feedback-driven automated fabrication of consistently thin cryo-lamellae

Cryo-electron tomography of 100-200 nm lamellae prepared by focused ion beam milling enables structural analysis of macromolecules within cells, tissues, and small organisms, but lamella preparation remains time-consuming and limits throughput. Although automated workflows enable unsupervised fabrication, they commonly produce variable results because fixed, pre-calibrated parameters are applied without accounting for lamella-to-lamella variability. To overcome these limitations, we developed an Adaptive Milling approach that adapts milling parameters based on imaging feedback. We implemented this approach for the critical polishing stage as Adaptive Polishing, which uses machine-learning-based segmentation of scanning electron microscopy images to identify lamella features predictive of imminent collapse. These are used to periodically assess whether to continue or stop thinning during polishing. We performed a systematic comparison of the quality of lamellae prepared with Adaptive Polishing and conventional automation, and demonstrate that Adaptive Polishing produces thinner lamellae with greater consistency in thickness, and performs well for in situ cryo-electron tomography. Adaptive Polishing was tested on three different microscopes with minimal model retraining required, demonstrating transferability of the approach. We implemented Adaptive Polishing as an open-source plugin for fibsemOS together with machine-learning model training software and pre-trained models.

molecular biology↗

bioRxiv2026

Systematic Evaluation of Feature Representations for Cancer-Associated sORF Prediction in Non-coding RNA

Short open reading frames (sORFs) within non-coding RNAs (ncRNAs) have arisen as a hidden layer of gene regulation, encoding small peptides that represent a new class of cancer regulators with diagnostic and therapeutic potential. However, inferring associations between sORFs to specific cancer types remains challenging and requires computational approaches for accurate prediction. Recently, the CoraL framework introduced the first computational approach for predicting cancer-associated peptides, focusing primarily on model architecture while overlooking how feature extraction strategies influence predictive accuracy. We present a systematic evaluation of machine learning models and feature extraction approaches to predict cancer-associated sORFs across 15 cancer types. We benchmarked seven traditional machine learning algorithms combined with three feature extraction methods: k-mer frequency, Word2Vec embeddings, and genomic language model (gLM)-based embeddings. To our knowledge, this is the first study applying gLM-derived embeddings to the prediction of cancer-associated sORFs in ncRNA. Our results show that traditional machine learning models with appropriate feature extraction outperform the CoraL baseline across all cancer types, achieving up to 10% higher accuracy in some of the 15 evaluated datasets. Interestingly, k-mer features consistently outperformed gLM embeddings without fine-tuning, suggesting that local sequence composition may provide more discriminative information for this task and that pre-trained genomic representations may require task-specific adaptation to fully capture these patterns. Additionally, we observed that the way sequences are tokenized, such as the k-mer length, can affect performance: longer fragments (e.g., k=7) sometimes reduced accuracy for Random Forest but had a smaller effect on MLP. Our findings suggest that appropriate feature engineering can provide greater improvements than increasing model complexity.

bioinformatics↗

bioRxiv2026

Benchmarking siRNA Prediction: The Role of Representation and Validation Strategies

Small interfering RNAs (siRNAs) offer transformative potential for targeted therapeutics, yet the design of highly effective and non-toxic candidates is hindered by the risk of off-target effects and RNA instability. A critical flaw in in silico prediction models is pervasive data leakage in cross-validation protocols, which artificially inflates performance metrics and produces untrustworthy results. To address this, we developed a rigorous framework that eliminates data leakage through strict cross-validation, leverages z-curves (3D representations of RNA physico-chemical properties) for context-aware sequence encoding, and identifies key sequence regions critical for efficacy. Our model achieves an AUC of 0.845 on leakage-free validation, surpassing prior work at 380x faster computation speed, demonstrating that superior representation trumps model complexity. Crucially, we demonstrate how experimental variability and cross-validation choices directly impact model reliability, establishing the first benchmarked methods for robust siRNA efficacy prediction. This work provides a foundation for trustworthy sequence design and validation in RNA therapeutics.

bioinformatics↗

bioRxiv2024

Depositing biological segmentation datasets FAIRly

Segmentation of biological images identifies regions of an image which correspond to specific features of interest, which can be analysed quantitatively to answer biological questions. This task has long been a barrier to conducting large-scale biological imaging studies as it is time- and labour-intensive. Modern artificial intelligence segmentation tools can automate this process, but require high quality segmentation data for training, which is challenging to acquire. Biological segmentation data has been produced for many years, but this data is not often reused to develop new tools as it is hard to find, access, and use. Recent disparate efforts (Iudin, et al., 2023; Xu, et al., 2021; Vogelstein, et al., 2018; Ermel, et al., 2024) have been made to facilitate deposition and re-use of these valuable datasets, but more work is needed to increase re-usability. In this work, we review the current state of publicly available annotation and segmentation datasets and make specific recommendations to increase re-usability following FAIR (findable, accessible, interoperable, re-usable) principles (Wilkinson, et al., 2016) for the future.

bioinformatics↗

bioRxiv2024

Generative Approaches for Nucleotide Sequences to Enhance Non-coding RNA Classification

Classifying non-coding RNA (ncRNA) sequences, particularly mirtrons, is essential for elucidating gene regulation mechanisms. However, the prevalent class imbalance in ncRNA datasets presents significant challenges, often resulting in overfitting and diminished generalization in machine learning models. In this study, GENerative Approaches for NUcleotide Sequences (GENNUS) is proposed, introducing novel data augmentation strategies using Generative Adversarial Networks (GANs) and Synthetic Minority Over-sampling Technique (SMOTE) to enhance ncRNA classification performance. Our GAN-based methods effectively generate high-quality synthetic data that capture the intricate patterns and diversity of real mirtron sequences, eliminating the need for extensive feature engineering. Through four experiments, it is demonstrated that models trained on a combination of real and GAN-generated data improve classification accuracy compared to traditional SMOTE techniques or only with real data. Our findings reveal that GANs enhance model performance and provide a richer representation of minority classes, thus improving generalization capabilities across various machine learning frameworks. This work highlights the transformative potential of synthetic data generation in addressing data limitations in genomics, offering a pathway for more effective and scalable ncRNA classification methodologies.

bioinformatics↗

bioRxiv2024

CHiMP: Deep Learning Tools Trained on Protein Crystallisation Micrographs to Enable Automation of Experiments

A group of three deep learning tools, referred to collectively as CHiMP (Crystal Hits in My Plate) were created for analysis of micrographs of protein crystallisation experiments at the Diamond Light Source (DLS) synchrotron, UK. The first tool, a classification network, assigns images into categories relating to experimental outcomes. The other two tools are networks that perform both object detection and instance segmentation, resulting in masks of individual crystals in the first case, and masks of crystallisation droplets in addition to crystals in the second case, allowing positions and sizes of these entities to be recorded. Creation of these tools used transfer learning, where weights from a pre-trained deep learning network were used as a starting point and re-purposed by further training on a relatively small set of data. Two of the tools are now integrated at the VMXi macromolecular crystallography beamline at DLS where they absolve the need for any user input both for monitoring crystallisation experiments and for triggering in situ data collections. The third is being integrated into the XChem fragment-based drug discovery screening platform, also at DLS, to allow automatic targeting of acoustic compound dispensing into crystallisation droplets.

biophysics↗

bioRxiv2023

A comparison of clustering models for inference of T cell receptor antigen specificity

The vast potential sequence diversity of TCRs and their ligands has presented an historic barrier to computational prediction of TCR epitope specificity, a holy grail of quantitative immunology. One common approach is to cluster sequences together, on the assumption that similar receptors bind similar epitopes. Here, we provide an independent evaluation of widely used clustering algorithms for TCR specificity inference, observing some variability in predictive performance between models, and marked differences in scalability. Despite these differences, we find that different algorithms produce clusters with high degrees of similarity for receptors recognising the same epitope. Our analysis highlights an unmet need for improvement of complex models over a simple Hamming distance comparator, and strengthens the case for use of clustering models in TCR specificity inference.

immunology↗

bioRxiv2022

Okapi-EM - a napari plugin for processing and analysing cryogenic serial FIB/SEM images

An emergent volume electron microscopy (vEM) technique called cryogenic serial plasma focused ion beam milling scanning electron microscopy (pFIB/SEM) can decipher complex biological structures by building a three-dimensional picture of biological samples at mesoscale resolution. This is achieved by collecting consecutive SEM images after successive rounds of FIB milling that expose a new surface after each milling step. Due to instrumental limitations, some image processing is necessary before 3D visualisation and analysis of the data is possible. SEM images are affected by noise, drift, and charging effects, that can make precise 3D reconstruction of biological features difficult. This paper presents Okapi-EM, an open-source Napari plugin(1) developed to process and analyse cryogenic serial FIB/SEM images. Okapi-EM enables automated image registration of slices, evaluation of image quality metrics specific to FIB-SEM imaging, and mitigation of charging artefacts. Implementation of Okapi-EM within the Napari framework ensures that the tools are both user- and developer-friendly, through provision of a graphical user interface and access to Python programming. Napari also hosts a variety of other image processing plugins so Okapi-EM tools can be integrated into and combined with other workflows. Okapi-EM can be downloaded freely at https://github.com/rosalindfranklininstitute/okapi-em, or installed from Python package index (PyPI). Impact statementCryogenic serial pFIB/SEM is an emerging microscopy technique that is used to visualise 3D structures of biological features at mesoscale resolutions(2). This technique requires common post processing of data such as alignment and charge mitigation to enable robust segmentation and analysis. In addition, approaches are needed to quantify data quality to enable an assessment of features and tune data acquisition parameters to enable optimal image acquisition. This article presents Okapi-EM, a combination of software tools designed to facilitate these important initial steps in assessing and processing images from these experiments. These tools have been assembled as a plugin for a popular 3D biological image visualiser called Napari, making their usage user-friendly and readily accessible.

bioinformatics↗

bioRxiv2022

Ot2Rec: A Semi-Automatic, Extensible, Multi-Software Tomographic Reconstruction Workflow

Electron cryo-tomography (cryo-ET) is an imaging technique for probing 3D structures with at the nanometre scale. This technique has been used extensively in the biomedical field to study the complex structures of proteins and other macromolecules. With the advancement in technology, microscopes are currently capable of producing images amounting to terabytes of data per day, posing great challenges for scientists as the speed of processing of the images cannot keep up with the ever-higher throughput of the microscopes. Therefore, automation is an essential and natural pathway on which image processing - from individual micrographs to full tomograms - is developing. In this paper, we present Ot2Rec, an open-source pipelining tool which aims to enable scientists to build their own processing workflows in a flexible and automatic manner. The basic building blocks of Ot2Rec are plugins which follow a unified API structure, making it simple for scientists to contribute to Ot2Rec by adding features which are not already available. In this paper, we also present three case studies of image processing using Ot2Rec, through which we demonstrate the speedup of using a semi-automatic workflow over a manual one, the possibility of writing and using custom (prototype) plugins, and the flexibility of Ot2Rec which enables the mix-and-match of plugins. We also demonstrate, in the supplementary information, a built-in reporting feature in Ot2Rec which aggregates the metadata from all process being run, and output them in the Jupyter Notebook and/or HTML formats for quick review of image processing quality. Ot2Rec can be found at https://github.com/rosalindfranklininstitute/ot2rec. Impact StatementThe field of cryo electron tomography has grown substantially in recent years, bringing about new advances in hardware and software which enable visualisation of cell and tissue architecture and proteins found in their native context. These same advances have, in some ways, stratified the field into those with access and those without. On the software side, this has emphasised the need for open-source options that do not require high levels of computational literacy to access. Additionally, it has highlighted the need for ways to both mix-and-match software for easy prototyping and comparisons between parameters and methods. Ot2Rec addresses these needs through a simple, unified plugin structure allowing the addition of existing software or the development of new and does so in a way which democratises access.

bioinformatics↗

bioRxiv2022

Cryo-plasma FIB/SEM volume imaging of biological specimens

Serial focussed ion beam scanning electron microscopy (FIB/SEM) enables imaging and assessment of sub-cellular structures on the mesoscale (10 nm to 10 m). When applied to vitrified samples, serial FIB/SEM is also a means to target specific structures in cells and tissues while maintaining constituents hydration shells for in-situ structural biology downstream. However, the application of serial FIB/SEM imaging of non-stained cryogenic biological samples is limited due to low contrast, curtaining and charging artefacts. We address these challenges using a cryogenic plasma FIB/SEM (cryo-pFIB/SEM). We evaluated the choice of plasma ion source and imaging regimes to produce high quality SEM images of a range of different biological samples. Using an automated workflow we produced three dimensional volumes of bacteria, human cells, and tissue, and calculated estimates for their resolution, typically achieving 20 to 50 nm. Additionally, a tag-free tool is needed to drive the application of in situ structural biology towards tissue. The combination of serial FIB/SEM with plasmabased ion sources promises a framework for targeting specific features in bulk-frozen samples (>100 m) to produce lamella for cryogenic electron tomography.

cell biology↗

bioRxiv2020

A massively multi-scale approach to characterising tissue architecture by synchrotron micro-CT applied to the human placenta

Multi-scale structural assessment of biological soft tissue is challenging but essential to gain insight into structure-function relationships of tissue/organ. Using the human placenta as an example, this study brings together sophisticated sample preparation protocols, advanced imaging, and robust, validated machine-learning segmentation techniques to provide the first massively multi-scale and multi-domain information that enables detailed morphological and functional analyses of both maternal and fetal placental domains. Finally, we quantify the scale-dependent error in morphological metrics of heterogeneous placental tissue, estimating the minimal tissue scale needed in extracting meaningful biological data. The developed protocol is beneficial for high-throughput investigation of structure-function relationships in both normal and diseased placentas, allowing us to optimise therapeutic approaches for pathological pregnancies. In addition, the methodology presented is applicable in characterisation of tissue architecture and physiological behaviours of other complex organs with similarity to the placenta, where an exchange barrier possesses circulating vascular and avascular fluid spaces.

bioengineering↗