Konstantin Avchaciov
Konstantin Avchaciov has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Peter O. Fedichev
GERO PTE. LTD., 133 Cecil Street 14-01 Keck Seng Tower, Singapore 069535
7 shared publications - Kai Nordlund
University of Helsinki
3 shared publications - Andrei V. Gudkov
Roswell Park Comprehensive Cancer Center, Buffalo, NY 14263, USA
2 shared publications Ekaterina L. Andrianova
Vaika, Inc., East Aurora, NY, USA
2 shared publications- Flyura Djurabekova
University of Helsinki
2 shared publications - Karsten Albe
Technical University of Darmstadt
2 shared publications
Latest Publications
- Open Access
HARVEST: Unlocking the Dark Bioactivity Data of Pharmaceutical Patents via Agentic AI
Pharmaceutical patents contain vast Structure–Activity Relationship tables documenting protein– ligand binding data. While technically public, this information remains computationally inaccessible and effectively dark, trapped in bulky documents that no existing database has systematically captured. We present HARVEST, a multi-agent large language model pipeline that autonomously extracts structured bioactivity records from USPTO patent archives at $0.11 per document. Applied to 164,877 patents, HARVEST produced 3.15 million activity records, recovering 326,342 unique scaffolds and 967 protein targets absent from BindingDB. This pipeline completed in under a week a task that would otherwise require over 55 years of continuous expert labor. Automated extraction achieves 80% agreement with human curated corpus of US patents from BindingDB, a conservative lower bound given identified errors within the reference data. We further introduce H-Bench, a structurally guaranteed held-out benchmark built from this recovered data. Evaluation of the leading open-source model Boltz-2 on H-Bench reveals a two-dimensional generalization gap: performance degrades both on novel scaffolds and on uncharacterized protein targets, exposing fundamental limitations of models trained on existing public repositories.
Open AccessHARVEST: Unlocking the Dark Bioactivity Data of Pharmaceutical Patents via Agentic AI
Pharmaceutical patents contain vast Structure–Activity Relationship tables documenting protein– ligand binding data. While technically public, this information remains computationally inaccessible and effectively dark, trapped in bulky documents that no existing database has systematically captured. We present HARVEST, a multi-agent large language model pipeline that autonomously extracts structured bioactivity records from USPTO patent archives at $0.11 per document. Applied to 164,877 patents, HARVEST produced 3.15 million activity records, recovering 326,342 unique scaffolds and 967 protein targets absent from BindingDB. This pipeline completed in under a week a task that would otherwise require over 55 years of continuous expert labor. Automated extraction achieves 80% agreement with human curated corpus of US patents from BindingDB, a conservative lower bound given identified errors within the reference data. We further introduce H-Bench, a structurally guaranteed held-out benchmark built from this recovered data. Evaluation of the leading open-source model Boltz-2 on H-Bench reveals a two-dimensional generalization gap: performance degrades both on novel scaffolds and on uncharacterized protein targets, exposing fundamental limitations of models trained on existing public repositories.
- Open Access
Thermodynamic scaling of canine aging and reversible clock deceleration by a reverse transcriptase inhibitor
Aging was tracked in a cohort of 99 “retired” sled dogs over four years to characterize the latent dynamics of physiological decline. Animals were randomized to receive either placebo or the reverse transcriptase inhibitor lamivudine for ∼30 months. We employed a variational autoregressive model to integrate longitudinal blood parameters and DNA methylation (DNAm) profiles. The model defines Biological Age (BA) as a signature of irreversible damage with Poissonian statistics, a feature conserved across mammalian scales. Lamivudine modulated age-independent latent variables and temporarily decelerated the DNAm clock in females, but these effects were reversible upon treatment discontinuation and did not alter the long-term BA trajectory. Critically, we show that physiological fluctuations are governed by a single systemic factor – an effective or phenotypic temperature representing an emergent (macroscopic) property. We show that while the rate of damage accumulation (the BA slope) is independent of this temperature, actuarial aging parameters (initial mortality and the Gompertz exponent) are strongly temperature dependent. This supports a model where mortality arises from effective activation across a protective free energy barrier that erodes with age. These findings identify phenotypic temperature as an important control variable governing the kinetics of organism-level failure, offering a compelling target for interventions aimed at extending healthspan by “squaring” the survival curve.
Open AccessThermodynamic scaling of canine aging and reversible clock deceleration by a reverse transcriptase inhibitor
Aging was tracked in a cohort of 99 “retired” sled dogs over four years to characterize the latent dynamics of physiological decline. Animals were randomized to receive either placebo or the reverse transcriptase inhibitor lamivudine for ∼30 months. We employed a variational autoregressive model to integrate longitudinal blood parameters and DNA methylation (DNAm) profiles. The model defines Biological Age (BA) as a signature of irreversible damage with Poissonian statistics, a feature conserved across mammalian scales. Lamivudine modulated age-independent latent variables and temporarily decelerated the DNAm clock in females, but these effects were reversible upon treatment discontinuation and did not alter the long-term BA trajectory. Critically, we show that physiological fluctuations are governed by a single systemic factor – an effective or phenotypic temperature representing an emergent (macroscopic) property. We show that while the rate of damage accumulation (the BA slope) is independent of this temperature, actuarial aging parameters (initial mortality and the Gompertz exponent) are strongly temperature dependent. This supports a model where mortality arises from effective activation across a protective free energy barrier that erodes with age. These findings identify phenotypic temperature as an important control variable governing the kinetics of organism-level failure, offering a compelling target for interventions aimed at extending healthspan by “squaring” the survival curve.
- Open Access
ProtoBind-Diff: A Structure-Free Diffusion Language Model for Protein Sequence-Conditioned Ligand Design
Designing small molecules that selectively bind to protein targets remains a central challenge in drug discovery. While recent generative models leverage 3D structural data to guide ligand generation, their applicability is limited by the sparsity and bias of structural resources. Here, we introduce ProtoBind-Diff, a structure-free masked diffusion model that conditions molecular generation directly on protein sequences via pre-trained language model embeddings. Trained on over one million active protein–ligand pairs from BindingDB, ProtoBind-Diff generates chemically valid, novel, and target-specific ligands without requiring structural supervision. In extensive benchmarking against structure-based models, ProtoBind-Diff performs competitively in docking and Boltz-1 evaluations and generalizes well to challenging targets, including those with limited training data. Despite never observing 3D information during training, its attention maps align with predicted binding residues, suggesting the model learns spatially meaningful interaction priors from sequence alone. This sequence-conditioned generation framework may unlock ligand discovery across the full proteome, including orphan, flexible, or rapidly emerging targets for which structural data are unavailable or unreliable.
Open AccessProtoBind-Diff: A Structure-Free Diffusion Language Model for Protein Sequence-Conditioned Ligand Design
Designing small molecules that selectively bind to protein targets remains a central challenge in drug discovery. While recent generative models leverage 3D structural data to guide ligand generation, their applicability is limited by the sparsity and bias of structural resources. Here, we introduce ProtoBind-Diff, a structure-free masked diffusion model that conditions molecular generation directly on protein sequences via pre-trained language model embeddings. Trained on over one million active protein–ligand pairs from BindingDB, ProtoBind-Diff generates chemically valid, novel, and target-specific ligands without requiring structural supervision. In extensive benchmarking against structure-based models, ProtoBind-Diff performs competitively in docking and Boltz-1 evaluations and generalizes well to challenging targets, including those with limited training data. Despite never observing 3D information during training, its attention maps align with predicted binding residues, suggesting the model learns spatially meaningful interaction priors from sequence alone. This sequence-conditioned generation framework may unlock ligand discovery across the full proteome, including orphan, flexible, or rapidly emerging targets for which structural data are unavailable or unreliable.
- Open Access
- 22 April 2025
- Published by Wiley in Aging Cell
- Vol. 24 (7) , e70060
AI ‐Driven Identification of Exceptionally Efficacious Polypharmacological Compounds That Extend the Lifespan of Caenorhabditis elegans
Analysis of existing lifespan‐extending geroprotective compounds suggested that polypharmacological compounds are the most effective geroprotectors, specifically those that bind multiple biogenic amine receptors. To test this hypothesis, we used graph neural networks to predict polypharmacological geroprotectors and evaluated them in Caenorhabditis elegans . Over 70% of the selected compounds extended lifespan, with effect sizes in the top 5% compared to all geroprotectors recorded in the DrugAge database. Thus, our study reveals that rationally designing polypharmacological compounds enables the design of geroprotectors with exceptional efficacy.
- 22 April 2025
- Published by Wiley in Aging Cell
- Vol. 24 (7) , e70060
Open AccessAI ‐Driven Identification of Exceptionally Efficacious Polypharmacological Compounds That Extend the Lifespan of Caenorhabditis elegans
Analysis of existing lifespan‐extending geroprotective compounds suggested that polypharmacological compounds are the most effective geroprotectors, specifically those that bind multiple biogenic amine receptors. To test this hypothesis, we used graph neural networks to predict polypharmacological geroprotectors and evaluated them in Caenorhabditis elegans . Over 70% of the selected compounds extended lifespan, with effect sizes in the top 5% compared to all geroprotectors recorded in the DrugAge database. Thus, our study reveals that rationally designing polypharmacological compounds enables the design of geroprotectors with exceptional efficacy.
- Open Access
Multiple Targets, One Goal: Compounding life‐extending effects through Polypharmacology
Analysis of lifespan‐extending compounds suggested the most effective geroprotectors target multiple biogenic amine receptors. To test this hypothesis, we used graph neural networks to predict such polypharmacological compounds and evaluated them in C. elegans . Over 70% of the selected compounds extended lifespan, with effect sizes in the top 5% compared to the DrugAge database. This reveals that rationally designing polypharmacological compounds enables the design of geroprotectors with exceptional efficacy. Key takeaways: The most effective known geroprotectors act by polypharmacological mechanisms. Graph neural networks predicted polypharmacological geroprotectors with a hit rate of 70%. The predicted polypharmacological geroprotectors are exceptionally effective. The predicted polypharmacological mechanism was experimentally confirmed. Rationally designing polypharmacological compounds results in geroprotectors with exceptional efficacy.
Open AccessMultiple Targets, One Goal: Compounding life‐extending effects through Polypharmacology
Analysis of lifespan‐extending compounds suggested the most effective geroprotectors target multiple biogenic amine receptors. To test this hypothesis, we used graph neural networks to predict such polypharmacological compounds and evaluated them in C. elegans . Over 70% of the selected compounds extended lifespan, with effect sizes in the top 5% compared to the DrugAge database. This reveals that rationally designing polypharmacological compounds enables the design of geroprotectors with exceptional efficacy. Key takeaways: The most effective known geroprotectors act by polypharmacological mechanisms. Graph neural networks predicted polypharmacological geroprotectors with a hit rate of 70%. The predicted polypharmacological geroprotectors are exceptionally effective. The predicted polypharmacological mechanism was experimentally confirmed. Rationally designing polypharmacological compounds results in geroprotectors with exceptional efficacy.
- Open Access
- 1 November 2022
- Published by Springer Nature in Nature Communications
- Vol. 13 (1) , 1-14
Unsupervised learning of aging principles from longitudinal data
Age is the leading risk factor for prevalent diseases and death. However, the relation between age-related physiological changes and lifespan is poorly understood. We combined analytical and machine learning tools to describe the aging process in large sets of longitudinal measurements. Assuming that aging results from a dynamic instability of the organism state, we designed a deep artificial neural network, including auto-encoder and auto-regression (AR) components. The AR model tied the dynamics of physiological state with the stochastic evolution of a single variable, the “dynamic frailty indicator” (dFI). In a subset of blood tests from the Mouse Phenome Database, dFI increased exponentially and predicted the remaining lifespan. The observation of the limiting dFI was consistent with the late-life mortality deceleration. dFI changed along with hallmarks of aging, including frailty index, molecular markers of inflammation, senescent cell accumulation, and responded to life-shortening (high-fat diet) and life-extending (rapamycin) treatments.
- 1 November 2022
- Published by Springer Nature in Nature Communications
- Vol. 13 (1) , 1-14
Open AccessUnsupervised learning of aging principles from longitudinal data
Age is the leading risk factor for prevalent diseases and death. However, the relation between age-related physiological changes and lifespan is poorly understood. We combined analytical and machine learning tools to describe the aging process in large sets of longitudinal measurements. Assuming that aging results from a dynamic instability of the organism state, we designed a deep artificial neural network, including auto-encoder and auto-regression (AR) components. The AR model tied the dynamics of physiological state with the stochastic evolution of a single variable, the “dynamic frailty indicator” (dFI). In a subset of blood tests from the Mouse Phenome Database, dFI increased exponentially and predicted the remaining lifespan. The observation of the limiting dFI was consistent with the late-life mortality deceleration. dFI changed along with hallmarks of aging, including frailty index, molecular markers of inflammation, senescent cell accumulation, and responded to life-shortening (high-fat diet) and life-extending (rapamycin) treatments.
- Open Access
Quantitative Characterization of Biological Age and Frailty Based on Locomotor Activity Records
We performed a systematic evaluation of the relationships between locomotor activity and signatures of frailty, morbidity, and mortality risks using physical activity records from the 2003 – 2006 National Health and Nutrition Examination Survey (NHANES) and UK BioBank (UKB). We proposed a statistical description of the locomotor activity tracks and transformed the provided time series into vectors representing physiological states for each participant. The Principal Components Analysis of the transformed data revealed a winding trajectory with distinct segments corresponding to subsequent human development stages. The extended linear phase starts from 35 40 years old and is associated with the exponential increase of mortality risks according to the Gompertz mortality law. We characterized the distance traveled along the aging trajectory as a natural measure of biological age and demonstrated its significant association with frailty and hazardous lifestyles, along with the remaining lifespan and healthspan of an individual. The biological age explained most of the variance of the log-hazard ratio that was obtained by fitting directly to mortality and the incidence of chronic diseases. Our findings highlight the intimate relationship between the supervised and unsupervised signatures of the biological age and frailty, a consequence of the low intrinsic dimensionality of the aging dynamics.
Open AccessQuantitative Characterization of Biological Age and Frailty Based on Locomotor Activity Records
We performed a systematic evaluation of the relationships between locomotor activity and signatures of frailty, morbidity, and mortality risks using physical activity records from the 2003 – 2006 National Health and Nutrition Examination Survey (NHANES) and UK BioBank (UKB). We proposed a statistical description of the locomotor activity tracks and transformed the provided time series into vectors representing physiological states for each participant. The Principal Components Analysis of the transformed data revealed a winding trajectory with distinct segments corresponding to subsequent human development stages. The extended linear phase starts from 35 40 years old and is associated with the exponential increase of mortality risks according to the Gompertz mortality law. We characterized the distance traveled along the aging trajectory as a natural measure of biological age and demonstrated its significant association with frailty and hazardous lifestyles, along with the remaining lifespan and healthspan of an individual. The biological age explained most of the variance of the log-hazard ratio that was obtained by fitting directly to mortality and the incidence of chronic diseases. Our findings highlight the intimate relationship between the supervised and unsupervised signatures of the biological age and frailty, a consequence of the low intrinsic dimensionality of the aging dynamics.
- 20 November 2015
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 118 (19) , 194306
Conditions for forming composite carbon nanotube-diamond like carbon material that retain the good properties of both materials
Carbon nanotubes are of wide interest due to their excellent properties such as tensile strength and electrical and thermal conductivity, but are not, when placed alone on a substrate, well resistant to mechanical wear. Diamond-like carbon (DLC), on the other hand, is widely used in applications due to its very good wear resistance. Combining the two materials could provide a very durable pure carbon nanomaterial enabling to benefit from the best properties of both carbon allotropes. However, the synthesis of high-quality diamond-like carbon uses energetic plasmas, which can damage the nanotubes. From previous works it is neither clear whether the quality of the tubes remains good after DLC deposition, nor whether the DLC above the tubes retains the high sp3 bonding fraction. In this work, we use experiments and classical molecular dynamics simulations to study the mechanisms of DLC formation on various carbon nanotube compositions. The results show that high-sp3-content DLC can be formed provided the deposition conditions allow for sidewards pressure to form from a substrate close beneath the tubes. Under optimal DLC formation energies of around 40–70 eV, the top two nanotube atom layers are fully destroyed by the plasma deposition, but layers below this can retain their structural integrity.
- 20 November 2015
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 118 (19) , 194306
Conditions for forming composite carbon nanotube-diamond like carbon material that retain the good properties of both materials
Carbon nanotubes are of wide interest due to their excellent properties such as tensile strength and electrical and thermal conductivity, but are not, when placed alone on a substrate, well resistant to mechanical wear. Diamond-like carbon (DLC), on the other hand, is widely used in applications due to its very good wear resistance. Combining the two materials could provide a very durable pure carbon nanomaterial enabling to benefit from the best properties of both carbon allotropes. However, the synthesis of high-quality diamond-like carbon uses energetic plasmas, which can damage the nanotubes. From previous works it is neither clear whether the quality of the tubes remains good after DLC deposition, nor whether the DLC above the tubes retains the high sp3 bonding fraction. In this work, we use experiments and classical molecular dynamics simulations to study the mechanisms of DLC formation on various carbon nanotube compositions. The results show that high-sp3-content DLC can be formed provided the deposition conditions allow for sidewards pressure to form from a substrate close beneath the tubes. Under optimal DLC formation energies of around 40–70 eV, the top two nanotube atom layers are fully destroyed by the plasma deposition, but layers below this can retain their structural integrity.
- 1 December 2014
- Published by Elsevier in Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms
- Vol. 341, 22-26
Effect of ion irradiation on structural properties of Cu64Zr36 metallic glass
- 1 December 2014
- Published by Elsevier in Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms
- Vol. 341, 22-26
Effect of ion irradiation on structural properties of Cu64Zr36 metallic glass
- 6 May 2013
- Published by AIP Publishing in Applied Physics Letters
- Vol. 102 (18)
Controlled softening of Cu64Zr36 metallic glass by ion irradiation
We study the effect of irradiation with 5–20 keV recoils on the topological and chemical short-range order of Cu64Zr36 metallic glass using molecular dynamics simulations. We show that within the cascade region, the structural backbone of stiff Cu-centered icosahedral units is destroyed, leading to locally softened areas. Under mechanical load, the formation of shear transformation zones is thus promoted in the damaged area. Our results suggest that irradiation is a means to introduce nucleation sites for multiple shear bands and thus prevents catastrophic failure due to the presence of a single critical shear band.
- 6 May 2013
- Published by AIP Publishing in Applied Physics Letters
- Vol. 102 (18)
Controlled softening of Cu64Zr36 metallic glass by ion irradiation
We study the effect of irradiation with 5–20 keV recoils on the topological and chemical short-range order of Cu64Zr36 metallic glass using molecular dynamics simulations. We show that within the cascade region, the structural backbone of stiff Cu-centered icosahedral units is destroyed, leading to locally softened areas. Under mechanical load, the formation of shear transformation zones is thus promoted in the damaged area. Our results suggest that irradiation is a means to introduce nucleation sites for multiple shear bands and thus prevents catastrophic failure due to the presence of a single critical shear band.