Research map / Probabilistic and scientific modeling / Molecules and proteins
SynLaD: Latent Diffusion for Generating Synthesizable Molecules Conditioned on 3D Pharmacophore Profiles
Miruna Cretu, John Bradshaw, Patricia Suriana, Saeed Saremi, Omar Mahmood, Kirill Shmilovich, Kangway V. Chuang, Vishnu Sresht, Colin A Grambow
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add joint 3D-decoder training with the synthesis head | 3D decoder latent diffusion · Unconditional generation | AiZynthFinder success: 0.59 -> 0.72 | Improved |
| add joint 3D training to the synthesis decoder | synthesis decoder · Unconditional generation | AiZynthFinder success: 0.81 -> 0.87 | Improved |
| use beam search (b=5) instead of sampling (ablation) | synthesis decoder · Autoencoder reconstruction | synthesis match rate (%): 43.4 -> 63.4 | Improved |
| use beam width 5 instead of 1 (ablation) | synthesis decoder · Autoencoder reconstruction | synthesis match rate (%): 47.6 -> 63.4 | Improved |
| increase beam width from 5 to 10 (ablation) | synthesis decoder · Autoencoder reconstruction | synthesis match rate (%): 63.4 -> 59.5 | Hurt |
| raise 3D loss weight lambda from 0.5 to 0.92 (ablation) | autoencoder · Autoencoder loss weights | 3D molecule match rate (%): 82.1 -> 98.5 | Improved |
| raise 3D loss weight lambda from 0.92 to 0.98 (ablation) | autoencoder · Autoencoder loss weights | building block match rate (%): 82.5 -> 80.6 | Hurt |
| raise KL weight beta from 1e-5 to 1e-4 (ablation) | autoencoder · Autoencoder loss weights | 3D molecule match rate (%): 98.5 -> 80.6 | Hurt |
| use latent dim 16 instead of 8 (ablation) | autoencoder, beam search · Autoencoder reconstruction | synthesis match rate (%): 58.4 -> 63.4 | Improved |
| use 100 sampling steps instead of 50 (ablation) | latent DiT · Unconditional generation | validity: 0.85 -> 0.89 | Improved |
| add classifier-free guidance (w=4.0) (ablation) | SynLaD synthesis outputs · Lit-PCBA hit diversification | hits (avg.): 17.9 -> 11.7 | Hurt |
| add classifier-free guidance (w=0.5) (ablation) | SynLaD · Lit-PCBA hit diversification | validity: 0.49 -> 0.56 | Improved |
Closest papers by meaning
- ShEPhERD: Diffusing shape, electrostatics, and pharmacophores for bioisosteric drug design (ICLR 2025 · similarity 0.90)
- Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generation (ICLR 2024 · similarity 0.90)
- Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · similarity 0.89)
- Towards Unified and Lossless Latent Space for 3D Molecular Latent Diffusion Modeling (NeurIPS 2025 · similarity 0.89)
- DecompOpt: Controllable and Decomposed Diffusion Models for Structure-based Molecular Optimization (ICLR 2024 · similarity 0.89)
- Chain-of-Generation: Progressive Latent Diffusion for Text-Guided Molecular Design (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
