Controllable Molecule Generation via Sparse Representation Editing: An Interpretability-Driven Perspective

An advanced computational framework for molecular design and optimization via interpretability-driven controllable molecule generation

Local Control

Generation of molecules containing specific atoms or functional groups at designated sites

COOH → Benzene Ring

COOH
→
Benzene

COOH → N

COOH
→
N

COOH → O

COOH
→
O

Benzene Ring → COOH

Benzene
→
COOH

Benzene Ring → N

Benzene
→
N

Benzene Ring → O

Benzene
→
O

N → COOH

N
→
COOH

N → Benzene Ring

N
→
Benzene

N → O

N
→
O

O → COOH

O
→
COOH

O → Benzene Ring

O
→
Benzene

O → N

O
→
N

Global Control

Customizing the overall structural and physicochemical properties within defined ranges

Lipophilicity Amplification

Lower Lipophilicity
→
Higher Lipophilicity

Lipophilicity Suppression

Higher Lipophilicity
→
Lower Lipophilicity

Ring System Amplification

Single Ring
→
Double Rings

Ring System Suppression

Multiple Rings
→
Fewer Rings

Aromaticity Amplification

Aliphatic
→
Aromatic

Aromaticity Suppression

Aromatic
→
Reduced Aromaticity

Bioisosteric Editing

Optimizing molecular properties via bioisosteric substitution while maintaining structural similarity

Molecular editing routes for water solubility improvement

Original
Step 1
Step 2
Step 3
Step 4
Step 5

Molecular editing routes for lipophilicity improvement

Original
Step 1
Step 2
Step 3
Step 4
Step 5

Editing Routes Planning

Peripheral and Skeletal editing of indoles with fluoroalkyl N-triftosylhydrazones

Peripheral editing of indoles with fluoroalkyl N-triftosylhydrazones

Original
Step 1
Step 2
Step 3
Step 4
Step 5

Skeletal editing of indoles with fluoroalkyl N-triftosylhydrazones

Original
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
Step 8

Editing Routes Optimization

Strategic optimization of molecular editing routes under synthetic feasibility constraints

Improving water solubility

Original
Step 1
Step 2

Increase the number of rotatable bonds

Original
Step 1
Step 2

Increase the number of sp³ carbon atoms

Original
Step 1
Step 2

Multi-Constraint Molecule Generation

Generating and optimizing drug candidates under complex constraints

Complex Constraint Optimization for Oral Drug Design

$1 <$ LogP $< 3$, molecular weight (MW) $< 500$, topological polar surface area (TPSA) $\leq 140$, H-bond donors $\leq 5$ and acceptors $\leq 10$, aromatic rings $\leq 3$, synthetic accessibility (SA) score $< 5$, and the quantitative estimate of drug-likeness (QED) $> 0.7$
Original
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
Step 8

Multi-Target Drug Discovery

GSK3$\beta$ activity $\geq 0.5$, JNK3 activity $\geq 0.5$, QED $> 0.6$, and SA score $< 4$
Original
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
Step 8
Step 9
Step 10
Step 11