A tumor small molecule computer virtual screening method based on global biological time sequence potential field

Abstract

The application discloses a kind of tumor small molecule computer virtual screening method based on global biological time series potential field, belong to the technical field of computing biology and GPU molecular simulation hardware.This application constructs tumor special global biological time series potential field, and distinguishes tumor driver mutation and random mutation by carrying causal deduction equation set on GPU hardware;Small molecules are simulated in batch by using multi-path subunit parallel architecture;Seventeen tumor biochemical indicators coupling system is built to evaluate molecular efficacy, toxicity and drug resistance risk;Optimal simulation precision is adaptively matched by power-efficacy constraint equation;Hardware weight is self-calibrated by relying on simulation error gradient descent.This application solves the pain points of traditional tumor virtual screening, such as high cost, poor target point distinguishing ability, large power consumption, toxicity risk postposition, and no adaptive optimization capability of model, and can quickly output high-activity, low-toxicity, low-drug-resistance anti-tumor candidate small molecules, suitable for high-throughput virtual screening scenarios of pharmaceutical companies and CRO institutions.

Description

Technical Field

[0001] This invention belongs to the fields of computational biology, computer virtual simulation of small molecules targeting solid tumors, and GPU high-throughput molecular simulation hardware technology. High-throughput computer-aided virtual simulation screening of small molecule compounds targeting pathogenic targets of various solid tumors such as lung cancer, liver cancer, and breast cancer, relying on GPUs. The simulation hardware can fully complete the entire chain of target causal localization, molecular binding simulation, organ toxicity and drug resistance risk control, and adaptive scheduling of simulation accuracy. Road data processing. Background Technology

[0002] The applicant had previously submitted a publicly available general mathematical deduction framework entitled "A Cross-Domain Digital Intelligence Interpretation and Creation Method Based on a Dual-Series Formula," but... This general framework is only designed for cross-domain quantization scenarios involving general text and images, and cannot adapt to dynamic conformations of biological macromolecules or time-series drug concentrations in organs. Complex biochemical simulation scenarios such as coupling and tumor mutation pathways lack dedicated biomedical field models, multi-layered risk control for toxicity and drug resistance, and drug simulation algorithms. Targeted improvements such as force scheduling and multi-dimensional biochemical indicator coupling calculations are insufficient to support the business needs of virtual screening of small molecules for solid tumors. Current oncology... The molecular simulation hardware and supporting algorithms have six major drawbacks: New drug activity verification is highly dependent on offline wet experiments, which are costly and time-consuming; relying solely on molecular... Surface matching cannot distinguish between tumor-driven mutations and random mutations, resulting in poor interpretability; the lack of a standardized biological potential gradient model makes accurate solution difficult. Optimal binding sites for molecular targets; high-throughput screening with high hardware costs due to the lack of computational adaptive scheduling; no prior hepatotoxicity or drug resistance risk control for candidate drugs. Molecular clinical trials suffer from numerous shortcomings; hardware employs unidirectional open-loop computation, simulation errors cannot correct for evaluation weights, and screening accuracy cannot be continuously improved. In summary, the industry... There is an urgent need for GPU hardware with features such as unified biological coupling temporal field, hierarchical pathogenesis pathway deduction, multi-index coupling evaluation, and adaptive computing power scheduling. A gradient self-calibrated closed-loop virtual screening scheme for small tumor molecules. Summary of the Invention

[0003] (I) Purpose of the invention: To address the shortcomings of existing tumor molecular simulation hardware in terms of weak mutation discrimination capability, lack of unified biological coupling temporal field, and matching of computing power. Addressing the shortcomings of rigidity, limited evaluation dimensions, delayed toxicity risk assessment, and lack of parameter-based self-optimization loops, this paper builds an adaptive system based on prior general mathematical deduction approaches. A tumor-specific global biological temporal potential field depicting the dynamic changes of biological macromolecules; employing a hierarchical GPU simulation sub-unit parallel architecture; and utilizing causal inference. Equations were used to analyze tumor-driving pathways; a coupled evaluation system for seventeen tumor biochemical indicators was established; and optimal simulations were solved based on drug efficacy and computational constraints. True precision; utilizing simulation error gradients to update indicator weights in reverse, forming a GPU hardware self-calibration closed loop, reducing the cost of developing anti-tumor small molecules. Mistakes in cost and cycle.

[0004] (II) Mathematical Formulas and Definitions of Basic Terms: 1. Global Biological Temporal Potential Field BioField_{tumor}(t)=F(PDB(\vec{x}),Mol(t),Org(t)), characterizes the temporal coupling relationship of target protein, small molecule, and organ drug concentrations; 2. A stratified pathogenic pathway causal deduction equation set, using the difference between two sets of biofields to distinguish between driver mutations and random mutations; 3. A weighted scoring formula. Score = \sum_{i=1}^{n} w_i \cdot S_i , weight 0.01 \le w_i \le 2.0 ; 4. Computing power cost function C_{tumor}(\rho)=a\rho^2+b\rho+c, and the drug efficacy quality function Q_{tumor}(\rho)=k\rho+m, where \rho∈[0,1] represents the simulation refinement. Combining computational power budgeting to solve for the optimal solution; 5. Lightweight mechanism: High-frequency millisecond calculation of dynamic efficacy indicators, and batch updating of slowly changing indicators such as liver and kidney toxicity. New, with preset templates for lung cancer, liver cancer, and breast cancer, and the ability to turn off irrelevant indicators to reduce GPU load.

[0005] (III) Core Technology Principles: 1. Unified modeling of the temporal potential field of the entire biological domain, using unified multi-dimensional biological data as the mathematical basis for parallel computing; 2. Causal deduction distinguishes tumor-causing mutations, improving the interpretability of target screening; 3. Gradient extremum solution finds the optimal molecular binding site, eliminating ineffective small molecules. 4. Coupled scoring of seventeen biochemical indicators to achieve comprehensive assessment of drug activity and risk; 5. Parallel computing of six types of GPU subunits to improve efficiency. 6. Small molecule library batch processing efficiency; 7. Computational power-fineness adaptive scheduling, balancing prediction accuracy and hardware power consumption; 8. Three-level risk control verification of hepatotoxicity and nephrotoxicity. Drug resistance, dosage, and interception of high-risk molecules; 8. Iteratively update weights based on simulation error gradient to build a hardware self-calibration closed loop.

[0006] (IV) Core Innovations: 1. Pioneering GPU-adapted tumor-wide biological temporal potential field, solving the fragmentation problem of simulation modules; 2. Hardware Built-in causal inference equations solve the problems of difficulty in distinguishing mutations and insufficient simulation interpretability; 3. A seventeen-item comprehensive biochemical assessment system breaks through the limitations of single... 1. Limitations of affinity scoring; 4. Tumor-specific computational optimization model, reducing hardware costs for high-throughput screening; 5. GPU six-subunit parallel architecture, large... Significantly improves batch screening speed; 6. Complete hardware self-calibration closed loop, allowing for rapid model adaptation when switching cancer datasets.

[0007] (V) Beneficial effects: Precisely identifies tumor pathogenic targets; proactively intercepts highly toxic and drug-resistant small molecules, reducing wet experiment input; equivalent The GPU computing power cost is reduced by 40% in terms of prediction accuracy; it supports one-click switching and adaptation for multiple solid tumors; and low-end GPUs can stably complete high-throughput screening. Simulation data can directly guide offline drug synthesis experiments.

[0008] (vi) Example: Taking EGFR lung cancer small molecule screening as an example, a GPU simulation server was used to import the mutation dataset and target protein. White structure, small molecule library and computational cost budget; hardware completes biofield initialization, parallel simulation, causal pathway analysis, and weighted scoring of seventeen indicators. A complete process including multi-layered risk control, optimal precision solution, and weighted iteration. Compared to traditional simulation methods, the detection rate of effective anti-tumor molecules is improved by 27%. Invalid simulation data was reduced by 32%, and the overall screening cycle was shortened by 45%. (vii) Description of the attached drawings: Figure 1 This is a flowchart of the entire execution process of the GPU hardware computer virtual simulation of tumor small molecules during the implementation of this invention. The flowchart shows the complete operation process from tumor biological data access, global biological time-series potential field initialization, hierarchical causal inference, target gradient simulation, multi-biochemical index coupling evaluation, tumor risk control, gradient weight iteration and simulation result output. It also shows the closed-loop iterative path of biochemical index weight parameter feedback.

[0010] Figure 2This is a schematic diagram of the critical determination curve of the biopotential gradient between molecules and tumor targets. The diagram includes two curves: the non-mutation baseline binding potential field and the EGFR mutation coupling potential field. The optimal binding critical point is marked, and the embedded small plot shows the changing trend of GPU gradient calculation, which is used to determine the critical conditions for the binding of small molecules and targets.

[0011] Figure 3 A weighted score bar chart for seventeen tumor biochemical indicators is provided. The bar chart displays the normalized simulation scores of core indicators such as target affinity, hepatotoxicity, nephrotoxicity, drug resistance probability, synthesis difficulty, and formulation stability. The remaining eleven indicators are marked with ellipses to present the quantitative evaluation results of each indicator for small molecule screening.

[0012] Figure 4 The graph shows the comparison curves of drug efficacy and GPU computing power consumption under different simulation refinement conditions. The solid line represents the change of the overall drug efficacy score with simulation refinement, and the dashed line represents the change of GPU computing power consumption with simulation refinement. The optimal simulation refinement value is marked in the graph to select simulation parameters that balance the screening effect and the computational cost.

Claims

1. A computer-based virtual screening method for small tumor molecules based on a global biological temporal potential field, mounted on GPU molecular simulation hardware, characterized in that, Includes the following steps: S1 Tumor Multi-Source Biological Data Hardware Import Steps Import solid tumor mutation gene datasets, target protein three-dimensional structures, open-source small molecule compound libraries, tumor ADMET toxicity assessment criteria, and hardware global computing power budget limit B through GPU simulation-supporting data reading units. S2 Tumor-Specific Biofield Hardware Initialization Steps The GPU loads the hierarchical pathogenesis pathway causal inference equation set, biological target gradient solution equation set, and biological temporal evolution equation set, and maps all imported biological data to the normalized simulation precision \rho∈[0,1] to complete the hardware simulation data model initialization. S3 Tumor Basic Constraint Hardware Initial Screening Steps The GPU loads the tumor-specific screening threshold inequality to limit the minimum binding potential of the target, the standardized molecular weight, the lipid-water partition coefficient, and the cytotoxicity red line, and batches out small molecule simulation data that do not meet the basic conditions. S4 Tumor Biofield Stratification Scoring Steps The GPU relies on the global biological temporal potential field to complete the small molecule-target coupling simulation, and divides the basic drug efficacy scores into three levels: strong, medium and weak, based on the molecular binding strength. S5 Multi-biochemical Indicator Weighted Correction Steps The GPU retrieves the weighted vector of tumor-specific biochemical indicators and performs a weighted summation of individual scores. If a molecule can simultaneously bind to multiple pathogenic targets, a preset multi-target coupling gain coefficient is superimposed to correct the overall score. S6 Multi-layered Tumor Risk Hardware Risk Control Interception Steps The GPU sequentially performs three levels of data risk control calculations: liver and kidney toxicity threshold verification, drug resistance mutation counterfactual inference, and clinical dosage determination. If any risk threshold is triggered, the current candidate small molecule simulation data will be eliminated. S7 Optimal Simulation Fineness Hardware Solution Steps The GPU is substituted into the tumor computing cost and drug efficacy quality constraint equations, and the optimal simulation refinement is solved with the hardware computing budget B as the constraint. The next batch of small molecules is then screened using the high-throughput virtual screening. S8 Simulation Result Output and Hardware Parameter Iteration Steps The GPU peripheral outputs a list of highly active and low-toxicity candidate molecules, a target binding mechanism report, and an organ drug resistance risk assessment report based on simulation predictions. All simulation biases from this round are collected to construct a loss function, and gradient descent is used to update the weights of biochemical indicators, completing a round of hardware simulation data self-calibration.

2. A computer-based virtual screening system for small tumor molecules based on a global biological temporal potential field, mounted on GPU-based tumor molecular simulation hardware, characterized in that, It includes nine hardware computing units: (1) Tumor multi-source biological data access unit; (2) Global biological temporal potential field GPU initialization unit; (3) Hierarchical pathogenic pathway causal deduction calculation unit; (4) Biological target gradient simulation calculation unit; (5) Weighted and coupled evaluation unit with multiple biochemical indicators; (6) Tumor multi-layered toxicity drug resistance risk control unit; (7) GPU simulation computing power adaptive scheduling unit; (8) Gradient descent hardware weight iteration unit; (9) Simulation result visualization peripheral output unit.

3. The computer-based virtual screening method for small tumor molecules based on a global biological temporal potential field according to claim 1, characterized in that, A total of seventeen tumor-specific biochemical indicators were used for comprehensive and multi-dimensional evaluation, including: target binding affinity, tumor inhibitory efficacy, hepatotoxicity threshold, nephrotoxicity threshold, oral bioavailability, tumor tissue penetration ability, synthesis difficulty, synthesis yield, administration convenience, anti-tumor adverse reactions, drug resistance mutation probability, target literature corroboration, molecular modification extensibility, simulation computing power overhead, tumor cell penetration rate, off-target probability, and in vivo stability of the formulation.

4. The computer-based virtual screening method for small tumor molecules based on a global biological temporal potential field according to claim 1, characterized in that, The GPU simulation hardware configuration features a lightweight mechanism for time-sharing sampling and scene template pruning: dynamic efficacy indicators are calculated at high frequency in milliseconds, while slowly changing toxicity indicators are updated at low frequency according to simulation batches; irrelevant indicators can be manually turned off based on computer virtual simulation scenarios of lung cancer, liver cancer, and breast cancer to reduce the real-time computing load on the hardware.

5. The computer-based virtual screening method for small tumor molecules based on a global biological temporal potential field according to claim 1, characterized in that, The causal inference equation set of the stratified pathogenic pathway uses the counterfactual difference comparison operation of the biofield to distinguish between tumor-driven mutations and random meaningless mutations.

6. The computer-based virtual screening method for small tumor molecules based on a global biological temporal potential field according to claim 1, characterized in that, The computing power cost function is a quadratic function C_{tumor}(\rho)=a\rho^2+b\rho+c, and the drug efficacy quality function is a linear function Q_{tumor}(\rho)=k\rho+m.

Images