Fig 1.
PavCA Index scores per days of conditioning for males and females.
The circles illustrate the mean Pavlovian Conditioning Approach (PavCA) Index scores obtained on each day of PavCA. The error bars depict difference- and correlation-adjusted 95% Confidence Intervals (CI), computed according to the method presented by Cousineau et al. [28].
Fig 2.
Distributions of PavCA Index scores across days of conditioning in the modeling sample.
Each bar represents frequency of observations in between intervals of Pavlovian Conditioning Approach (PavCA) Index scores. Each graphic contains 10 bars, a total of 189 subject scores were classified on each day of conditioning.
Table 1.
Classification of PavCA Index scores with the k-Means algorithm where k = 3.
Table 2.
Classification of PavCA Index scores (modeling sample) with the derivative method.
Fig 3.
Density function of mean PavCA Index scores and its derivative function.
The upper curve represents a density estimate of mean Pavlovian Conditioning Approach (PavCA) Index scores obtained from the last two days of conditioning, extracted with a smoothing function of the true distribution of scores. The lower curve depicts the first derivative of this density function. The blue line marks the local minimum of the derivative function, indicating a suitable cutoff value for goal-tracking scores. The yellow line indicates the local maximum of the derivative in an area associated to sign-tracking scores.
Fig 4.
Frequencies in each group according to the 3 classification methods.
( (A) Frequencies of subjects classified in each group/phenotype with 3 classification methods in the modeling sample. (B) Validation sample. Proportions are distorted the most with the ±0.5 cutoff in the validation sample, compared to the k-Means and derivative.
Fig 5.
Progression of scores across days of conditioning for each phenotype group according to the classification methods.
(A) K-Means results. (B) Derivative method. (C) Standard ±0.5 cutoff. Classification was performed on the mean scores of the last two days of conditioning in the modeling sample. Each curve shows phenotype groups progressing over time. The error bars illustrate difference- and correlation-adjusted 95% Confidence Intervals (CI), computed according to the method presented by Cousineau et al. [28]. Each method seems to capture similar subjects and trends, they become distinct at the edges of each phenotype.
Fig 6.
Classification from 1D vs. 3D k-Means trained on the pooled scores from Days 5 and 6 of the modeling sample.
(A) Pooled Pavlovian Conditioning Approach (PavCA) Index scores from Days 5 and 6, classified with the 1D k-Means (the algorithm was performed directly on the scores). (B) Same pooled PavCA Index scores classified with the 3D k-Means. The 3D method does not produce a linear delimitation, prompting for further research.
Fig 7.
The relationship between pooled PavCA Index scores from the last 2 days of conditioning and three behavioral parameters.
Pooled data (Days 5 and 6) from the modeling sample of the latency scores, lever presses and food cup entries represented in 3D and color-coded with their associated Pavlovian Conditioning Approach (PavCA) Index scores.
Fig 8.
Mean PavCA Index score distributions and classification values for each method.
(A) Distribution of mean Pavlovian Conditioning Approach (PavCA) Index scores from the last two days of conditioning in the modeling sample (n = 189). (B) Mean scores in the validation sample (n = 34). Colored lines indicate cutoff values determined by each method. Red lines illustrate the ± 0.5 cutoff values, yellow lines the k-Means cutoffs, green lines the derivative method. Yellow asterisks (*) illustrate the centroid value for k-Means, green asterisks illustrate the peak location of the density function (derivative method).