The current study describes the development of in silico models based on a novel alternative of the MTD-PLS methodology (Partial-Least-Squares variant of Minimal Topologic Difference) developed by our group to predict the inhibition of GSK-3β by indirubin derivatives. The new MTD-PLS methodology involves selection rules for the PLS equation coefficients based on physico-chemical considerations aimed at reducing the bias in the output information. These QSAR models have been derived using calculated fragmental descriptors relevant to binding including polarizability, hydrophobicity, hydrogen bond donor, hydrogen bond acceptor, volume and electronic effects. The MTD-PLS methodology afforded moderate but robust statistical characteristics (R2 Y(CUM) = 0.707, Q2(CUM) = 0.664). The MTD-PLS model obtained has been validated in terms of predictive ability by joined internal-external cross-validation applying Golbraikh-Tropsha criteria and Y-randomization test. The information supplied by the MTD-PLS model has been evaluated against Fujita-Ban outcomes that afforded a statistically reliable model (R2=0.923). Furthermore, the results originated from QSAR models were laterally validated with docking insights that suggested the substitution pattern for the design of new indirubins with improved pharmacological potential against GSK-3β. The new restriction rules introduced in this paper are applicable and provide reliable results in accordance with physico-chemical reality.