Assessment of fat depots for several
goat body parts is an expensive and time-consuming
task requiring a trained technician. Therefore, the establishment
of models to predict fat depots based on
data requiring simpler and easier procedures, such as
ultrasound measurements, that could be carried out in
vivo, would be a major advantage. An interesting alternative
to the use of multiple linear regression models is
the use of partial least squares or artificial neural network
models because they allow the establishment of
one model to simultaneously predict different fat depots
of interest. In this work, the applicability of these models
to simultaneously predict 7 goat fat depots (subcutaneous
fat, intermuscular fat, total carcass fat, omental
fat, kidney and pelvic fat, mesenteric fat, and total
body fat) was investigated. Although satisfactory correlation
and prediction results were obtained using the
multiple partial least squares model (cross-verification
and validation R2 and standard prediction error values
between 0.66 and 0.98 and 247 and 2,168, respectively),
the best global correlation and prediction performances
were achieved with the multiple radial basis function
artificial neural network (verification and validation R2
and standard prediction error values between 0.82 and
0.96 and 304 and 1,707, respectively). These 2 multiple
models allowed correlating and predicting simultaneously
the 7 goat fat depots based on the goat BW and
on only 2 ultrasonic measures (lumbar subcutaneous
fat between fifth and sixth vertebrae and the fat depth
at the third sternebra). Moreover, both multiple models
showed better results compared with those obtained
with multiple linear regression models proposed in previous
work.
