双药研究工作台Drug-pair research workbench

支持中文常用病名、英文病名或 EFO ID，最多 60 种；不足时不补造。Common Chinese disease names, English names or EFO IDs; up to 60 drugs, without fabricated entries.

保留疾病研究阶段与直接靶点关联，用于后续排序。缺失证据不补造；不使用活性阈值或方向假设。Disease development phase and direct target associations inform ranking. Missing evidence is not fabricated; activity thresholds and direction hypotheses are not used.

使用分子量、cLogP、TPSA、氢键供体、可旋转键，以及可用的 Morgan 指纹构造对称配对特征。Symmetric pair features use MW, cLogP, TPSA, H-bond donors, rotatable bonds and available Morgan fingerprints.

同条件成粒实验（成功=1，失败=0）训练回归树集成。至少 12 个不同配对并优于均值基线，才启用估计。成粒分不是校准概率；重复实验冲突时保留未知。A tree ensemble learns from same-condition assembly outcomes (success=1, failure=0). Estimates require at least 12 distinct pairs and improvement over the mean baseline. Scores are not calibrated probabilities; conflicting replicates remain unresolved.

更改实验条件会隔离数据；原有 0–1 质量记录保留，但不当作成粒标签。Conditions isolate datasets. Legacy 0–1 quality records are retained but never treated as formation labels.

H-RACS 是抗癌双药协同预测工具。本页参考其结构与细胞背景思路，未连接 H-RACS 服务或复现其训练模型。H-RACS predicts anti-cancer drug synergy. This page follows its structure-and-cell-context approach; it does not connect to the H-RACS service or reproduce its trained model.

当前本地模型仅使用配对分子特征，在相同细胞、指标、剂量方案与条件内学习。无基因表达或靶点网络输入；不能跨细胞推断。The local model uses molecular pair features only, within a fixed cell model, metric, dose protocol and condition group. It has no gene-expression or target-network inputs and cannot infer across cell models.

外部预测仅作为导入证据，不进入实验训练。游离双药协同不证明纳米制剂协同；H-RACS 分数阈值需按原模型定义设置。Imported predictions are evidence only, never experimental training labels. Free-drug synergy does not establish nanoformulation synergy. Set the H-RACS cutoff according to its original model definition.

将已取得的预测整理为模板 CSV；填写原始模型或结果链接。仅接受当前细胞与指标的结果。Map existing predictions to the template CSV and provide the original model or result URL. Only results matching the current cell model and metric are accepted.

在双阶段保留的配对中，结合疾病证据、协同分和结构多样性选择最多 20 种单药；每种至少保留一个入选搭档。不足时不补足。模型分歧不是置信区间，配对分组验证不能证明对全新药物有效。Up to 20 individual drugs are selected from pairs retained by both stages, using disease evidence, synergy scores and structural diversity. Each drug retains at least one selected partner; shortfalls are not filled. Model disagreement is not a confidence interval, and pair-grouped validation does not establish generalization to unseen drugs.

显示前 30 对First 30 pairs shown

默认值仅示例。任何条件变化会切换数据组；暂不支持跨条件预测。Defaults are examples. Changing any condition switches data groups; cross-condition prediction is not supported.