A scientist's day, in Affinity
You design a novel format
A program is building a bispecific T-cell engager against CD20 and CD3. The shape doesn't fit your team's existing antibody or scFv format catalog; it's a new format the program will run multiple constructs in.
The engineer defines the format on Affinity's format diagram canvas, then generates a panel of bispecific TCEs in the format the diagram describes by plugging in parts from the built-in catalog: the same variable domains, constant regions, linkers, and hinges the team's other engineered work is built from, not a free-text description standing in for them. What the engineer lays out once becomes a reusable format: every bispecific the program registers against it inherits the architecture, so the twenty constructs that follow are all the same defined shape rather than twenty hand-assembled one-offs.
You assemble an ADC
The next construct in the program is an ADC version of the lead anti-CD20 antibody, an antibody-payload conjugate with a cleavable linker carrying a tubulin-disrupting cytotoxic warhead.
The engineer first draws the cytotoxic warhead using the built-in chemical structure sketcher, then registers a panel of ADCs using different linkers to connect the payload to the lead antibody. Now registered as a conjugate, the ADC is a single composite the platform reasons about as one molecule, with the antibody, payload, and linker connected together in one record. The warhead's chemistry the antibody's sequence, even the line back to the source clone are all traceable from one place. The ADC isn't multiple records that a scientist has to remember are related; it's the molecule, whole, with its lineage intact.
You build a panel from engineered constructs
The program wants twenty bispecific constructs moving forward into characterization. The bispecific format defined earlier needs to be populated with specific variable-domain pairs from the team's anti-CD20 and anti-CD3 panels.
Now the format needs filling — twenty bispecifics, each a real pair of binders drawn from the program's anti-CD20 and anti-CD3 work. A panel is how the engineer turns the abstract format into those twenty constructs: the paired variable domains coming off clone identification slot into the format's arms, so what the panel holds isn't a list of picked clones but a set of assembled bispecifics in the format the program chose, traceable back to their origin out of discovery.
Which domains go in is a filtering question, and the engineer narrows the discovery output the way the decision actually runs — by binding affinity, expression yield, sequence-liability count, V-gene usage, which target a clone hits — or starts from a set the team already curated. The panel that comes out carries forward into expression, characterization, and reporting as one named thing that the rest of the program works against.
You compare formats side by side
The program needs to decide whether the lead anti-CD20 candidate should advance as an IgG, an scFv, or a bispecific T-cell engager — three different formats carrying the same set of underlying variable domains. The question isn't which single candidate is best; it's which format is best.
The engineer builds the same underlying domains into three panels — one IgG, one scFv, one bispecific — and scores each against the program's agreed profile: binding, expression, stability, liabilities, weighted the way the program decided. Ranked on a dashboard side by side, the three formats stop being a matter of opinion — the IgG panel's leaders sit next to the bispecific panel's leaders on the same criteria, and the format call is read off the data rather than argued to consensus.
Formats rarely survive the first pass intact, and revising one doesn't overwrite it: dropping the format that underperforms or adding a new variant arm spins off a derived panel that keeps its parent's lineage. Every format the program tried stays on the record — so months later, why did we drop the trispecific? is a question with an answer, not a memory.
