I don't know the academic literature around TypingDNA-like methods, but I can throw in some ideas. In essence what you're describing is the Android Pattern Unlock, but instead of a 3x3 grid, you have a, say, 200x200 pixel writing area, right?
If it was, the "password" space would be huge, correct?
Technically yes, but be careful not to conflate the theoretical number of possibilities with the distribution of what people actually select. For example:
90% of people's scribbles will be continuous strokes in the centre of the writing area rather than, say, doing random jumps right to the edges.
Fuzzy matching: two strokes by the same person will never be identical, so you will need an algorithm that accepts "similar" strokes as being equivalent. The fingerprint unlock chips in iPhone and Android have to deal with this, so you could do some reading there.
I don't have any numbers to do a calculation, but my intuition is that in practice this password space will end up being in the same ballpark as a 12 character password.
Capture and replay attacks: keyloggers are a problem with passwords; what's preventing a malicious program from logging your mouse movements and replaying them? You'll notice that RECATPCHA challenges are designed to be record&replay-proof. Fingerprint scanners deal with record&replay by giving the scanner chip its own CPU; on successful match it uses a cryptographic key to sign a message to the OS saying that you have successfully authenticated so that the OS and software never actually see the fingerprint -- nothing to record. I wonder what protection TypingDNA has against record&replay attacks.
Bottom line: Thinking through the attack surfaces is the most fun part of security work. I suspect this idea wouldn't give any more security than a traditional password, but it's neat and fun to think about!