DDMut-PPI: predicting effects of mutations on protein-protein interactions using graph-based deep learning.

Protein-protein interactions (PPIs) play a vital role in cellular functions and are essential for therapeutic development and understanding diseases. However, current predictive tools often struggle to balance efficiency and precision in predicting the effects of mutations on these complex interactions. To address this, we present DDMut-PPI, a deep learning model that efficiently and accurately predicts changes in PPI binding free energy upon single and multiple point mutations. Building on the robust Siamese network architecture with graph-based signatures from our prior work, DDMut, the DDMut-PPI model was enhanced with a graph convolutional network operated on the protein interaction interface. We used residue-specific embeddings from ProtT5 protein language model as node features, and a variety of molecular interactions as edge features. By integrating evolutionary context with spatial information, this framework enables DDMut-PPI to achieve a robust Pearson correlation of up to 0.75 (root mean squared error: 1.33 kcal/mol) in our evaluations, outperforming most existing methods. Importantly, the model demonstrated consistent performance across mutations that increase or decrease binding affinity. DDMut-PPI offers a significant advancement in the field and will serve as a valuable tool for researchers probing the complexities of protein interactions. DDMut-PPI is freely available as a web server and an application programming interface at https://biosig.lab.uq.edu.au/ddmut_ppi.

DDMut: predicting effects of mutations on protein stability using deep learning.

Understanding the effects of mutations on protein stability is crucial for variant interpretation and prioritisation, protein engineering, and biotechnology. Despite significant efforts, community assessments of predictive tools have highlighted ongoing limitations, including computational time, low predictive power, and biased predictions towards destabilising mutations. To fill this gap, we developed DDMut, a fast and accurate siamese network to predict changes in Gibbs Free Energy upon single and multiple point mutations, leveraging both forward and hypothetical reverse mutations to account for model anti-symmetry. Deep learning models were built by integrating graph-based representations of the localised 3D environment, with convolutional layers and transformer encoders. This combination better captured the distance patterns between atoms by extracting both short-range and long-range interactions. DDMut achieved Pearson's correlations of up to 0.70 (RMSE: 1.37 kcal/mol) on single point mutations, and 0.70 (RMSE: 1.84 kcal/mol) on double/triple mutants, outperforming most available methods across non-redundant blind test sets. Importantly, DDMut was highly scalable and demonstrated anti-symmetric performance on both destabilising and stabilising mutations. We believe DDMut will be a useful platform to better understand the functional consequences of mutations, and guide rational protein engineering. DDMut is freely available as a web server and API at https://biosig.lab.uq.edu.au/ddmut.

Large-Scale Observations Support Aboveground Vegetation as an Important Biological Mercury Sink in the Tibetan Plateau.

Mercury, a pervasive global pollutant, primarily enters the atmosphere through human activities and legacy emissions from the land and oceans. A significant portion of this mercury subsequently settles on land through vegetation uptake. Characterizing mercury storage and distribution within vegetation is essential for comprehending regional and global mercury cycles. We conducted an unprecedented large-scale aboveground vegetation mercury survey across the expansive Tibetan Plateau. We find that mosses (31.1 ± 0.5 ng/g) and cushion plants (15.2 ± 0.7 ng/g) outstood high mercury concentrations. Despite exceptionally low anthropogenic mercury emissions, mercury concentrations of all biomes exceeded at least one-third of their respective global averages. While acknowledging the role of plant physiological factors, statistical models emphasize the predominant impact of atmospheric mercury on driving variations in mercury concentrations. Our estimations indicate that aboveground vegetation on the plateau accumulates 32-12+21 Mg (interquartile range) mercury. Forests occupy the highest biomass and store 82% of mercury, while mosses, representing only 3% of the biomass, disproportionally contribute 13% to mercury storage and account for 43% (2.5-1.4+3.0 Mg/year) of annual mercury assimilation by vegetation. Additionally, our study underscores that extrapolating aboveground vegetation mercury storage from lower-altitude regions to the Tibetan Plateau can lead to substantial overestimation, inspiring further exploration in alpine ecosystems worldwide.

Joint genotypic and phenotypic outcome modeling improves base editing variant effect quantification.

CRISPR base editing screens enable analysis of disease-associated variants at scale; however, variable efficiency and precision confounds the assessment of variant-induced phenotypes. Here, we provide an integrated experimental and computational pipeline that improves estimation of variant effects in base editing screens. We use a reporter construct to measure guide RNA (gRNA) editing outcomes alongside their phenotypic consequences and introduce base editor screen analysis with activity normalization (BEAN), a Bayesian network that uses per-guide editing outcomes provided by the reporter and target site chromatin accessibility to estimate variant impacts. BEAN outperforms existing tools in variant effect quantification. We use BEAN to pinpoint common regulatory variants that alter low-density lipoprotein (LDL) uptake, implicating previously unreported genes. Additionally, through saturation base editing of LDLR, we accurately quantify missense variant pathogenicity that is consistent with measurements in UK Biobank patients and identify underlying structural mechanisms. This work provides a widely applicable approach to improve the power of base editing screens for disease-associated variant characterization.

Joint genotypic and phenotypic outcome modeling improves base editing variant effect quantification.

CRISPR base editing screens are powerful tools for studying disease-associated variants at scale. However, the efficiency and precision of base editing perturbations vary, confounding the assessment of variant-induced phenotypic effects. Here, we provide an integrated pipeline that improves the estimation of variant impact in base editing screens. We perform high-throughput ABE8e-SpRY base editing screens with an integrated reporter construct to measure the editing efficiency and outcomes of each gRNA alongside their phenotypic consequences. We introduce BEAN, a Bayesian network that accounts for per-guide editing outcomes and target site chromatin accessibility to estimate variant impacts. We show this pipeline attains superior performance compared to existing tools in variant classification and effect size quantification. We use BEAN to pinpoint common variants that alter LDL uptake, implicating novel genes. Additionally, through saturation base editing of LDLR, we enable accurate quantitative prediction of the effects of missense variants on LDL-C levels, which aligns with measurements in UK Biobank individuals, and identify structural mechanisms underlying variant pathogenicity. This work provides a widely applicable approach to improve the power of base editor screens for disease-associated variant characterization.

kinCSM: Using graph-based signatures to predict small molecule CDK2 inhibitors.

Protein phosphorylation acts as an essential on/off switch in many cellular signaling pathways. This has led to ongoing interest in targeting kinases for therapeutic intervention. Computer-aided drug discovery has been proven a useful and cost-effective approach for facilitating prioritization and enrichment of screening libraries, but limited effort has been devoted providing insights on what makes a potent kinase inhibitor. To fill this gap, here we developed kinCSM, an integrative computational tool capable of accurately identifying potent cyclin-dependent kinase 2 (CDK2) inhibitors, quantitatively predicting CDK2 ligand-kinase inhibition constants (pKi ) and classifying different types of inhibitors based on their favorable binding modes. kinCSM predictive models were built using supervised learning and leveraged the concept of graph-based signatures to capture both physicochemical properties and geometry properties of small molecules. CDK2 inhibitors were accurately identified with Matthew's Correlation Coefficients (MCC) of up to 0.74, and inhibition constants predicted with Pearson's correlation of up to 0.76, both with consistent performances of 0.66 and 0.68 on a nonredundant blind test, respectively. kinCSM was also able to identify the potential type of inhibition for a given molecule, achieving MCC of up to 0.80 on cross-validation and 0.73 on the blind test. Analyzing the molecular composition of revealed enriched chemical fragments in CDK2 inhibitors and different types of inhibitors, which provides insights into the molecular mechanisms behind ligand-kinase interactions. kinCSM will be an invaluable tool to guide future kinase drug discovery. To aid the fast and accurate screening of CDK2 inhibitors, kinCSM is freely available at https://biosig.lab.uq.edu.au/kin_csm/.

Structure-guided machine learning prediction of drug resistance mutations in Abelson 1 kinase.

Kinases play crucial roles in cellular signalling and biological processes with their dysregulation associated with diseases, including cancers. Kinase inhibitors, most notably those targeting ABeLson 1 (ABL1) kinase in chronic myeloid leukemia, have had a significant impact on cancer survival, yet emergence of resistance mutations can reduce their effectiveness, leading to therapeutic failure. Limited effort, however, has been devoted to developing tools to accurately identify ABL1 resistance mutations, as well as providing insights into their molecular mechanisms. Here we investigated the structural basis of ABL1 mutations modulating binding affinity of eight FDA-approved drugs. We found mutations impair affinity of type I and type II inhibitors differently and used this insight to developed a novel web-based diagnostic tool, SUSPECT-ABL, to pre-emptively predict resistance profiles and binding free-energy changes (ΔΔG) of all possible ABL1 mutations against inhibitors with different binding modes. Resistance mutations in ABL1 were successfully identified, achieving a Matthew's Correlation Coefficient of up to 0.73 and the resulting change in ligand binding affinity with a Pearson's correlation of up to 0.77, with performances consistent across non-redundant blind tests. Through an in silico saturation mutagenesis, our tool has identified possibly emerging resistance mutations, which offers opportunities for in vivo experimental validation. We believe SUSPECT-ABL will be an important tool not just for improving precision medicine efforts, but for facilitating the development of next-generation inhibitors that are less prone to resistance. We have made our tool freely available at http://biosig.unimelb.edu.au/suspect_abl/.