القائمة
Mitigating the effect of activity cliffs using Triplet Soft Margin (TSM) loss on High-Value Activity Cliff Triplets (HV-ACTs).
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Running molecular geometry optimization and molecular dynamics (MD) simulations using ASE and MLIPs.
Generating 3D molecular conformations and preparing input files for quantum chemistry DFT calculations.
Constructing Gaussian output neural networks and training deep ensembles to quantify aleatoric and epistemic uncertainty.
Implementing Delta-ML (residual learning between low and high levels of theory) and model transfer learning.
Optimizing molecular feature weights and performing feature selection via Differentiable Information Imbalance (DII).
Active learning loops for MLIPs using uncertainty-biased configuration selection and DFT query oracle simulation.
| name | activity_cliff_awareness |
| description | Mitigating the effect of activity cliffs using Triplet Soft Margin (TSM) loss on High-Value Activity Cliff Triplets (HV-ACTs). |
Use this skill when training models for quantitative structure-activity relationship (QSAR) or affinity prediction, where small structural changes can cause massive drops or jumps in biological activity (Activity Cliffs).
An Activity Cliff (AC) occurs when two molecules are structurally highly similar but have radically different target affinity. Standard machine learning models suffer on these cliff compounds because the representation layers naturally project similar structures near each other, smoothing out the target space.
The Activity Cliff Awareness (ACA) framework penalizes smooth representation mapping of cliff compounds. It constructs High-Value Activity Cliff Triplets (HV-ACTs):
The combined loss function is: $$\text{Loss}{\text{ACA}} = \text{Loss}{\text{Regression}} + \alpha \cdot \text{Loss}{\text{TSM}}$$ where $\text{Loss}{\text{TSM}}$ is the Triplet Soft Margin loss acting on representation embeddings $h$: $$\text{Loss}_{\text{TSM}} = \ln\left(1 + \exp(d(h_a, h_p) - d(h_a, h_n) + m)\right)$$ where $d$ is a distance metric (e.g., Euclidean) and $m$ is a soft margin.
Run the script activity_cliff_awareness.py to train a model incorporating TSM loss:
python scripts/activity_cliff_awareness.py --alpha 0.5 --margin 1.0 --epochs 30
This script mines triplets and demonstrates how representation distances change during training.