What kind of meat was a dinosaur? Predicting muscle composition from animal and muscle traits, tested by hiding whole species

The meat of a dinosaur cannot be measured, but the meat of living animals can, and its composition differs with the animal and with the muscle. We asked whether composition can be predicted from traits that a dinosaur also has, and tested this on living animals before applying it. We linked 225 living species, their muscles and 9,730 compiled composition values from four national food-composition tables, 22 studies of wild animals and a fibre-type dataset. Predictors were the animal's body mass, growth rate, thermoregulation, class, locomotion, habitat and diet, the muscle's body region, mechanical role and activity, and how each value was recorded. Under species-grouped nested cross-validation (hyperparameters tuned in inner folds, the tree method chosen with the outer-fold results, so the correlations are exploratory and selection-affected), tree ensembles predicted skeletal-muscle composition with correlations of 0.66 for total n-6 fatty acids, 0.60 for the slow-fibre share, 0.57 for DHA, 0.53 for saturated fatty acids, 0.52 for protein, 0.50 for zinc and 0.40 for fat, and when the heaviest 20% of species were hidden their errors stayed 2–59% below those of the training mean for all 12 components tested. Saturated fat followed locomotion (ruminants), DHA followed life in water, linoleic acid and ash followed body size, and the slow-fibre share followed the mechanical role of the muscle. Applied to 1,255 dinosaur species, theropod thigh was predicted fattier (3.8 vs 1.7–1.8 g/100 g) and richer in n-6 fatty acids than that of herbivorous lineages; the order of cuts in slow fibres depended on the body plan: the neck ranked highest in all 691 quadrupedal and facultatively bipedal species, whereas the thigh ranked highest in 447 of the 564 bipeds, with the neck highest in 101 small bipeds and the loin in 16 piscivore-labelled ones. Each lineage difference could be traced to the living animals it draws on, such as crocodiles and turtles for the DHA of ornithischians, poultry for the n-6 of theropods and humans for the slow fibres of bipeds. Clustering placed most dinosaurs with their own lineage; herbivorous therizinosaurs and piscivore-labelled theropods joined carnivorous theropods because shared traits outweighed diet, and among within-cluster pairs whose trait contributions differed by more than 1 SD unit in total, opposing contributions cancelled by a median 28%. In ornithischians and sauropodomorphs, predictions did not change with size beyond about 1 t.