Skip to main content
Figure 5 | BMC Systems Biology

Figure 5

From: The genotype-phenotype relationship in multicellular pattern-generating models - the neglected role of pattern descriptors

Figure 5

Predicting model parameters from solution characteristics. Top row: multivariate calibration illustrated. (a) Regression coefficient for predicting θD from sensory profiles, estimated by PLSR. While θN was kept constant at 0.1, pD and pN were likewise calibrated for. (b) Predicted θD vs. known θD in the calibration samples, based on cross-validation to avoid over-fitting. Middle row: prediction in a normal sample. (c) A solution image generated with P = [θD, θD, pD, pN] = [0.7, 0.1, 10, 3]. (d) Image reconstructed from predicted sensory parameters P = [0.71, 0.1, 9.43, 3.57]. Right: outlier tests (1, relative leverage; 2, residual variance) indicate a valid prediction. Bottom row: prediction in an abnormal sample. (e) An image generated with a parameter combination outside the range calibrated for, P = [0.7, 0.5, 10, 3]. (f) Image reconstructed from predicted sensory parameters P = [0.79, 0.1, 11.3, 1.7]. The reconstruction is obviously bad, as expected. But since outlier tests (right) were too high, the invalid prediction caused automatic outlier warning.

Back to article page