Biology subjects

Ayaz, H.

Publications and source records attributed to Ayaz, H..

5 recordsLinked to original sources

bioRxiv2026

When perfect is a warning: explanations expose shortcuts that accuracy hides in cancer gene-expression classifiers

When the classes in a training set come from different data sources, a classifier can score near-perfectly while learning where a sample came from. We show this for cancer classifiers trained on tumours from The Cancer Genome Atlas and normal tissue from the Genotype Tissue Expression project. Six model families, including a tabular foundation model, reached an area under the curve of 1.000 yet called 80 to 97% of tumour adjacent normal tissue cancer, and 100% in an independent cohort; a few matched normal samples per tissue removed most of this shortcut. Near-perfect accuracy could not tell a shortcut from an honestly easy task, flagging 87 to 92% of models trained on such tasks. We introduce a label free score, computed from model explanations, that uses the fact that a source difference looks the same in every tissue whereas cancer changes do not. It raised no false alarms, ranked models by their real reliance on the source, transferred to unseen model types and data, and failed where its theory predicts.

bioinformatics↗

bioRxiv2025

Predicting Future SARS-CoV-2 Mutations using Deep Learning

SARS-CoV-2 continues to spread over the world steadily as opposed to many earlier estimations that it would disappear in less than two years. Even though SARS-CoV-2 vaccines have reduced the speed of the infection significantly, they could not fully stop it. On the contrary, the World Health Organization has recently published cautionary statements that infection counts are on the rise, and a huge wave is expected in winter. Vaccines mostly target specific regions of the virus. The high mutation rate of SARS-CoV-2 is one essential tool that the virus exploits to escape from the available vaccines. Therefore, researchers have been working on designing next-generation vaccines against the new variants of the virus. Nevertheless, SARS-CoV-2 acquires new mutations faster than we can adapt our vaccines due to long clinical trial periods. Hence, there is a need for computational tools that can predict future SARS-CoV-2 mutations before they even emerge. In this paper, we propose several deep-learning-based methods to estimate the possible future mutations in SARS-CoV-2 genome. We design and evaluate various ensemble and bagging architectures enriched with a large set of genomic, biochemical, and phylogenetic features. We evaluate our models on the GISAID data and demonstrate that the best-performing methods achieve an F1-Macro score of 0.78.

bioinformatics↗

bioRxiv2020

Effects of prefrontal theta burst stimulation on neuronal activity and subsequent eating behavior: an interleaved rTMS and fNIRS study

The dorsolateral prefrontal cortex (dlPFC) and dorsomedial prefrontal cortex (dmPFC) are important nodes for self-control and decision-making, but through separable processes (cognitive control versus evaluative processing). This study aimed to examine the effects of excitatory brain stimulation (intermittent theta-burst stimulation; iTBS) targeting the dlPFC and dmPFC on food choice. iTBS was hypothesized to decrease consumption of appetitive snack foods, via enhanced interference control for dlPFC stimulation and reduced delay discounting for dmPFC stimulation. Using a single-blinded, between-subjects design, participants (N = 43) were randomly assigned to of the three conditions: 1) iTBS targeting the left dlPFC, 2) iTBS targeting bilateral dmPFC, or 3) sham. Participants then completed two cognitive tasks (delay discounting (DD) and Flanker), followed by a taste test. fNIRS imaging revealed increases in medial PFC activity were evident in the dmPFC stimulation group during the DD task; likewise, a neural efficiency effect was observed in the dlPFC stimulation group during the Flanker. Gender significantly moderated consumption during the taste test, with females in the dmPFC showing paradoxical increases in food consumption compared to sham. Findings are consistent with possible amplification of positive evaluative processing in the presence of dietary restraint, vis-a-vis excitation of the mPFC.

neuroscience↗

bioRxiv2020

Scalable Classification of Organisms into a Taxonomy Using Hierarchical Supervised Learners

Taxonomy of living organisms gains major importance in making the study of vastly heterogeneous living things easier. In addition, various fields of applied biology (e.g., agriculture) depend on classification of living creatures. Specific fragments of the DNA sequence of a living organism have been defined as DNA barcodes and can be used as markers to identify species efficiently and effectively. The existing DNA barcode-based classification approaches suffer from three major issues: (i) most of them assume that the classification is done within a given taxonomic class and/or input sequences are prealigned, (ii) highly performing classifiers, such as SVM, cannot scale to large taxonomies due to high memory requirements, (iii) mutations and noise in input DNA sequences greatly reduce the taxonomic classification accuracy. In order to address these issues, we propose a multi-level hierarchical classifier framework to automatically assign taxonomy labels to DNA sequences. We utilize an alignment-free approach called spectrum kernel method for feature extraction. We build a proof-of-concept hierarchical classifier with two levels, and evaluated it on real DNA sequence data from BOLD systems. We demonstrate that the proposed framework provides higher accuracy than regular classifiers. Besides, hierarchical framework scales better to large datasets enabling researchers to employ classifiers with high accuracy and high memory requirement on large datasets. Furthermore, we show that the proposed framework is more robust to mutations and noise in sequence data than the non-hierarchical classifiers.

bioinformatics↗

bioRxiv2016

Measuring speaker-listener neural coupling with functional near infrared spectroscopy

The present study investigates brain-to-brain coupling, defined as inter-subject correlations in the hemodynamic response, during natural verbal communication. We used functional near-infrared spectroscopy (fNIRS) to record brain activity of speakers telling stories and listeners comprehending audio recordings of these stories. Listeners brain activity was correlated with speakers with a delay. This between-brain correlation disappeared when verbal communication failed. We further compared the fNIRS and functional Magnetic Resonance Imaging (fMRI) recordings of listeners comprehending the same story and found a relationship between the fNIRS oxygenated-hemoglobin concentration changes and the fMRI BOLD in brain areas associated with speech comprehension. This correlation between fNIRS and fMRI was only present when data from the same story were compared between the two modalities and vanished when data from different stories were compared; this cross-modality consistency further highlights the reliability of the spatiotemporal brain activation pattern as a measure of story comprehension. Our findings suggest that fNIRS is a powerful tool for investigating brain-to-brain coupling during verbal communication. As fNIRS sensors are relatively low-cost and can even be built into wireless, portable, battery-operated systems, these results highlight the potential of broad utilization of this approach in everyday settings for augmenting communication and interaction.

neuroscience↗