Apparatus, systems and methods for determining, for each respective phenotypic characterization in a set of {T.sub.1, . . . , T.sub.k} characterizations, that a test specimen has the respective characterization are provided. A pairwise probability function g.sub.pq(X, W.sub.pq), for a phenotypic pair (T.sub.p, T.sub.q) in {T.sub.1, . . . , T.sub.k} is learned using a training population. W.sub.pq is a set of parameters derived from Y for (T.sub.p, T.sub.q) by substituting each Y.sub.1 in Y into g.sub.pq(X, W.sub.pq), as X, where Y.sub.i is the set of cellular constituent abundance values from sample i in the training population exhibiting T.sub.p or T.sub.q. The learning step is repeated for each (T.sub.p, T.sub.q) in {T.sub.1 . . . , T.sub.k}, thereby deriving pairwise probability functions G={g.sub.1,2(X, W.sub.1,2), . . . , g.sub.k-1, k(X, W.sub.k-1, k)}. Pairwise probability values P={p.sub.1,2, . . . , p.sub.k-1, k} are computed, where each p.sub.pq is equal to g.sub.pq(Z, W.sub.pq) in G, the probability that the test specimen has T.sub.p and not T.sub.q, where Z is cellular constituent abundance values of the test specimen.