${\mathcal{G}}_S$	Molecular graph of a chemical species S
$h_v,h_i^{\left(l\right)}$	Node (atom) feature vector; node feature at layer l
$N\left(v\right)$	Neighbour set of node v; $\mathrm{deg}(·)$ is node degree
$h^{\left(G\right)}$	Global GCNN embedding (molecular fingerprint)
$f_s,f_l,f_r$	Structural feature vectors of solvent, ligand, reducing agent
${\lambda }^{\prime },{\lambda }_k$	Candidate condition–feature vector; k-th original vector
${\mu }_W({\lambda }^{\prime }\mid {\lambda }_k)$	Membership of candidate ${\lambda }^{\prime }$ relative to ${\lambda }_k$
$n,d$	Number of original samples in the current training fold; number of features
$\mathcal{J}$	Index set of training-fold features with non-zero variance
D	Training-fold covariance matrix; diagonal in the principal AAE
$D^{-1}$	Precision matrix corresponding to D
${\sigma }_j^2$	Population variance of feature j within the current training fold
a	Minimum membership threshold (affi in the implementation)
$N_{\mathrm{aug}}$	Target number of retained training-fold virtual samples
s	Random seed controlling stochastic candidate generation and selection
$\alpha ,\beta$	Attribution deviation coefficients (far/near acceptance)
ACC, F1, MCC	Accuracy, F1 score, Matthews correlation coefficient