A stain is found at a crime scene. The expert takes a swab, and a few hours later the instrument outputs a list of proteins that made up the fluid. Previously, a rapid test could be wrong if multiple fluids were present: blood mixed with saliva, and the strip gave a false result. Now, a mass spectrometer weighs each molecule, and three algorithms translate these weights into the language of a forensic scientist.

The first interpreter knows individual words. The researchers selected 526 peptides (short protein fragments) that occur in only one of five fluids: blood, saliva, semen, urine, or vaginal fluid. When the instrument finds such a fragment, the interpreter confidently names the source. On pure samples, this approach never erred for blood and saliva, and for the other fluids it erred in fewer than one case out of ten.

The second interpreter watches the grammar. Instead of individual words, it compares how many of certain peptides there are relative to others. Blood has lots of hemoglobin, while saliva has almost none, but saliva has more amylase. The researchers selected 250 such pairwise ratios, and their balance reveals the composition of a mixture. This approach proved more accurate than the first on mixed stains: it correctly identified fluids in three out of four cases.

The third interpreter did not learn words in advance. It was given 1110 features selected by the ReliefF algorithm and asked to find patterns on its own. The random forest model built thousands of decision trees, and through their voting it learned to distinguish fluids. On pure samples, it never erred, and on mixtures it achieved 95.9% accuracy. Interestingly, these three interpreters use almost non-overlapping feature sets: only 15 peptides were common to all approaches. This means they complement each other, like three witnesses who saw the crime from different angles.

Today, such interpretation can already work in a forensic laboratory. It is especially important in sexual assault cases, where one must distinguish semen from skin cells, even if the stain has dried or mixed with other fluids. So far, the method has been tested on volunteers and pure samples, but the next step is trials on real stains from crime scenes, where proteins may have degraded due to time or heat. If the interpreters also handle that, forensic science will gain a reliable way to read traces, independent of the whims of rapid tests.