Biology subjects

Pak, M.

Publications and source records attributed to Pak, M..

3 recordsLinked to original sources

bioRxiv2025

A more comprehensive and reliable analysis of individual differences with generalized random forest for high-dimensional data: validation and guidelines

Analyzing individual differences in treatment or exposure effects is a central challenge in psychology and behavioral sciences. Conventional statistical models have focused on average treatment effects, overlooking individual variability, and struggling to identify key moderators. Generalized Random Forest (GRF) can predict individualized treatment effects, but current implementations suffer from two critical limitations: (1) prediction performances vary substantially across random initializations, and (2) identification of key moderator is limited in high-dimensional settings. Here, we introduce two methodological advances to address these issues. First, a seed ensemble strategy stabilizes predictions by aggregating models trained under different random initializations. Second, a backward elimination procedure systematically identifies key moderators from high-dimensional inputs. Simulation analyses across diverse scenarios demonstrate that our approach achieves reliable and valid predictions across random seeds, improved performance in moderator identification, and robust generalization to independent data. To facilitate adoption and interpretation, we provide step-by-step guidance using large-scale neuroimaging dataset (N = 8,778) with reusable code. These enhancements make GRF more reliable for modeling individual differences in treatment effects, supporting data-driven hypothesis generation, and identification of responsive subgroups.

neuroscience↗

bioRxiv2024

MULAN: Multimodal Protein Language Model for Sequence and Structure Encoding

Most protein language models (PLMs), which are used to produce high-quality protein representations, use only protein sequences during training. However, the known protein structure is crucial in many protein property prediction tasks, so there is a growing interest in incorporating the knowledge about the protein structure into a PLM. In this study, we propose MULAN, a MULtimodal PLM for both sequence and ANgle-based structure encoding. MULAN has a pre-trained sequence encoder and an introduced Structure Adapter, which are then fused and trained together. According to the evaluation on 7 downstream tasks of various nature, both small and medium-sized MULAN models show consistent improvement in quality compared to both sequence-only ESM-2 and structure-aware SaProt. Importantly, our model offers a cheap increase in the structural awareness of the protein representations due to finetuning of existing PLMs instead of training from scratch. We perform a detailed analysis of the proposed model and demonstrate its awareness of the protein structure. The implementation, training data and model checkpoints are available at https://github.com/DFrolova/MULAN.

bioinformatics↗

bioRxiv2022

Modulation of RNA binding properties of the RNA helicase UPF1 by its activator UPF2

The NMD helicase UPF1 is a prototype of the superfamily 1 (SF1) of RNA helicases that bind RNA with high affinity and translocate on it in an ATP-dependent manner. Previous studies showed that UPF1 has a low basal catalytic activity that is greatly enhanced upon binding of its interaction partner, UPF2. Activation of UPF1 by UPF2 entails a large conformational change that switches the helicase from an RNA-clamping mode to an RNA-unwinding mode. The ability of UPF1 to bind RNA was expected to be unaffected by this activation mechanism. Here we show, using a combination of biochemical and biophysical methods, that binding of UPF2 to UPF1 drastically reduces the affinity of UPF1 for RNA, leading to a release of the bound RNA. Although UPF2 is capable of binding RNA in vitro, our results suggest that dissociation of the UPF1-RNA complex is not a consequence of direct competition in RNA binding but rather an allosteric effect that is likely mediated by the conformational change induced by binding of the helicase to its activator. We discuss these results in light of transient interactions forged during mRNP assembly, particularly in the UPF1-dependent mRNA decay pathways.

biochemistry↗