Predictive model for determining changes in molecular state

A computer-implemented method, comprising:

• receiving a plurality of sequencing reads, each sequencing read comprising a sequence tag that uniquely identifies a parent cell-free DNA (cfDNA) fragment;
• grouping the sequencing reads into read families according to identical sequence tags and, for each read family, generating an error-corrected consensus sequence by aligning the reads of the read family and resolving per-base discrepancies that fail to satisfy a quality threshold;
• for each of a plurality of subjects, constructing a molecular state vector that comprises, for each of a plurality of genomic loci, a quantitative allele-fraction value derived from the error-corrected consensus sequences;
• storing, in a training datastore, a plurality of ordered pairs each including (i) a first molecular state vector captured at a first time point and (ii) a subsequent molecular state vector captured at a second time point later than the first time point;
• and
• training, with the plurality of ordered pairs, a predictive model that parameterizes state-transition probabilities between molecular state vectors conditioned on elapsed time, wherein the predictive model is configured to receive a query molecular state vector associated with an evaluation subject and compute, for each candidate subsequent molecular state vector, a probability that the evaluation subject will transition from the query molecular state vector to the candidate subsequent molecular state vector within a user-specified time window.