ANAnastasiia Nemashkalo

Anastasiia Nemashkalo

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Anastasiia Nemashkalo has a total of 32 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Small but Mighty: smaller protein language models outperform larger models for antibody thermostability prediction

    by Rudrasis Chakraborty, Jacob Pettit, Mary Silva, Kevin Mcloughlin, Emily A Lyon, Colin S Kruse, Corey Quackenbush, Anastasiia Nemashkalo, Yuliya Kunde, Emilia Solomon, et al.

    Predicting antibody thermostability from sequence is a central challenge in computational antibody design, with direct implications for manufacturability, formulation stability, and shelf-life. We systematically evaluate the utility of protein language model (pLM) embeddings for this task using two large-scale yeast surface display datasets: a naïve scFv library (128,268 sequences) and a designed Fab library (113,612 sequences). To ensure realistic generalization estimates, we adopt a germline-aware cross-validation strategy preventing the data leakage that arises when germline-conserved sequences are randomly split across training and test sets. We compare a broad set of pLMs, including multiple scales of ESM2 (8 M, 150 M, and 15B parameters), ProtBERT, ProtGPT2, and antibody-specific models (AbBERT, AbESM, AbLang2). Strikingly, ESM2-8 M achieves the best individual performance, matching or exceeding models up to 1900 × larger, suggesting that larger models are over-parameterized for this setting. Analysis of embedding distance correlations reveals that BERT-based models form a coherent cluster while AbLang2 captures complementary information; exploiting this complementarity via a stacking ensemble of ESM2-8 M and AbLang2 yields consistent further gains across both datasets. Performance variation across models is driven more strongly by dataset structure—particularly clade composition and sample size—than by model scale. Together, these findings establish that compact, general-purpose pLMs provide an effective and computationally efficient foundation for antibody thermostability prediction, and that strategic combination of architecturally distinct embeddings can further improve performance.

  • Open Access

    In Vitro Evolution and Computational Approaches to Predict, Prevent and Control Future Pandemics

    by Antonietta Maria Lillo, Anastasiia Nemashkalo Davis, Esen Sokullu, Corey Richard Quackenbush, Yuliya A. Kunde, Cesar Augusto Lopez Bautista, Kshitij Wagh

    The U.S. Department of Energy's Office of Scientific and Technical Information

  • Real-time, single-molecule observation of biomolecular interactions inside nanophotonic zero mode waveguides

    by A Nemashkalo, M E Phipps, S P Hennelly, P M Goodwin

    Living cells rely on numerous protein-protein, RNA-protein and DNA-protein interactions for processes such as gene expression, biomolecular assembly, protein and RNA degradation. Single-molecule microscopy and spectroscopy are ideal tools for real-time observation and quantification of nucleic acids-protein and protein-protein interactions. One of the major drawbacks of conventional single-molecule imaging methods is low throughput. Methods such as sequencing by synthesis utilizing nanofabrication and single-molecule spectroscopy have brought high throughput into the realm of single-molecule biology. The Pacific Biosciences RS2 sequencer utilizes sequencing by synthesis within nanophotonic zero mode waveguides. A number of years ago this instrument was unlocked by Pacific Biosciences for custom use by researchers allowing them to monitor biological interactions at the single-molecule level with high throughput. In this capability letter we demonstrate the use of the RS2 sequencer for real-time observation of DNA-to-RNA transcription and RNA-protein interactions. We use a relatively complex model–transcription of structured ribosomal RNA from E. coli and interactions of ribosomal RNA with ribosomal proteins. We also show evidence of observation of transcriptional pausing without the application of an external force (as is required for single-molecule pausing studies using optical traps). Overall, in the unlocked, custom mode, the RS2 sequencer can be used to address a wide variety of biological assembly and interaction questions at the single-molecule level with high throughput. This instrument is available for use at the Center for Integrated Nanotechnologies Gateway located at Los Alamos National Laboratory.


  • Low-defect-density ZnO homoepitaxial films grown by low-temperature ALD

    by David Elam, Eduardo Ortega, Anastasiia Nemashkalo, Yuri Strzhemechny, Arturo Ayon, Arturo Ponce, Andrey A. Chabanov

    We report atomic layer deposition (ALD) of ZnO thin films on O-polar surface crystalline ZnO substrates at the relatively low temperatures of 120, 150, and 200 °C. The as-grown ZnO films are studied with aberration-corrected transmission electron microscopy and diffraction contrast, photoluminescence (PL), and surface photovoltage (SPV) spectroscopy. We find that the homoepitaxial films have a monocrystalline structure with the density of basal stacking faults comparable to that of the substrate (∼1011 cm−2) and that the stacking faults can induce high lattice strain due to their interaction with the inversion domain boundaries. The narrow excitonic PL linewidth (2 meV at 8 K) and the sharp SPV bandgap transition confirm the high quality of the ZnO films. Despite similarities in the film properties, the growth temperature has an effect on the density and spatial distribution of intrinsic defects. Our results demonstrate a considerable potential of ALD ZnO homoepitaxy for fabricating high-quality ZnO nanostructures and attaining viable p-type ZnO.